-Earl_Babbie-_The_Practice_of_Social_Research(BookFi)

(1) In my work, I've seen that it's relatively easy to established techniques for quantitative data analysis are make some observations of social life and speculate powerful tools to use in that pursuit, the really powerful about the meaning of what has been observed. Unfortu- discoveries are never produced by the rote administra- nately, such speculation is unlikely to make much of a tion of techniques. contribution to our understanding of social life. (4) The most difficult task for social scientists lies in (2) Doing even asimplistic quantitative data analy- producing powerful analyses of qualitative data. This re- sis is more difficult, because it requires at least some quires the same dedication and ability discussed in (3); low-level statistical skills. All too often, however, we're however, qualitative analysis depends more on the indi- confronted with statistical data analyses that don't vidual insights of the researcher than on the tools avail- really mean much. Terms such as quantiphrenia and able to support the analysis. Qualitative analysis re- scientism have sometimes been used in reference to mains today as much an art as ascience. attempts to mimic the physical sciences without any true meaning. I hope the chapters that make up this part of the book will give you some of the tools and sharpen the (3) Doing sophisticated, meaningful quantitative inSights needed to produce sophisticated data analyses, data analyses requires much thought and imagination. whether qualitative or quantitative. It does not necessarily require high-powered statistics, however, as much of the work of Paul Lazarsfeld and Chapter 13 examines qualitative data analysiS. We'll Sam Stouffer shows. What's needed instead is the will- begin by examining some of the theoretical groundings ingness to search for, and the ability to recognize, mean- for this approach. Then we'll look at some conceptual ingful patterns among variables. Although the many procedures you should find useful in the search for meaning among qualitative data. Finally, I'll demonstrate 375

some of the computer programs that have been created the more commonly used statistical methods in social specifically for qualitative data analysis. science research, including an overview of some of the more advanced methods of multivariate analysis. Rather The first of several discussions on the logic of quan- than merely showing how to compute statistics by these titative data analysis is presented in Chapter 14. We'll methods (computers can do that), I've attempted to begin with an examination of methods of analyzing and place them in the context of earlier theoretical and logi- presenting data related to asingle variable. Then we'll cal discussions. Thus, you should come away from this turn to the relationship between two variables and learn chapter knowing when to use various statistical mea- how to construct and read simple percentage tables. sures as well as how to compute them. The chapter ends with apreview of multivariate analy- sis and adiscussion of sociological diagnostics. Finally, Chapter 17 addresses social research as lit- erature: how to read it and how to write it. The materi- Chapter 15 describes the elaboration model of data als of this chapter are essentially bookends for the re- analysis developed by Paul Lazarsfeld at Columbia Uni- search process: areview of the literature early in the versity. Here we'll look further into multivariate analysis. project involves the skills of reading social research, and Chapter 15 also presents the logic of causal analysis writing it comes into play in the communication of your through the use of percentage tables. We'll apply this results to others in the form of your research report. same logic when we use other statistical techniques in Chapter 16. This logical model was developed for use with quantitative data, but I think you'll see how appro- priate it is to reasoning with qualitative data as well.

Qualitative Data Analysis Introduction Computer Programs for Qualitative Data Linking Theory and Analysis Discovering Patterns Leviticus as Seen through Grounded Theory Method NUD*IST Semiotics Conversation Analysis Using NVivo to Understand Women Film Directors, Qualitative Data Processing by Sandrine Zerbib Coding Memoing The Qualitative Analysis Concept Mapping of Quantitative Data Sociology®Now~: Research Methods Use this online tool to help you make the grade on your next exam. After reading this chapter, go to the \"Online Study Resources\" at the end of the chapter for instructions on how to benefit from SodologyNoll': Research Methods.

378 Chapter 13: Qualitative Data Analysis Introduction theory. As a result, I've already talked about quali- tative data analysis in earlier discussions of field re- Later chapters in Part 4 of the book will deal vvith search and content analysis. In quantitative re- the qUailtitative analysis of social research data, search, it's sometimes easy to get caught up in the sometimes called statiSTical al1alysis. Recent decades logistics of data collection and in the statistical anal- of social science research have tended to focus on ysis of data, thereby losing sight of theory for a time. quantitative data analysis techniques. This focus, This is less likely in qualitative research, where data however. sometimes conceals another approach to collection, analysis, and theory are more intimately making sense of social observations: qualitative intertwined. analysis-methods for examining social research data without converting them to a numerical for- In the discussions to follow, we'll use the mat. This approach predates quantitative analysis. image of theory offered by Anselm Strauss and It remains a useful approach to data analysis and is Juliet Corbin (1994: 278) as consisting of \"plausible even enjoying a resurgence of interest among social relationships proposed among concepts and sets of scientists. concepTS.\" They stress \"plausible\" to indicate that theories represent our best understanding of how Although statistical analyses may intimidate life operates. The more our research confirms a some students, the steps involved can sometimes be particular set of relationships among particular learned in a rote manner. That is, with practice, the concepts, however. the more confident we become rote exercise of quantitative skills can produce an that our understanding corresponds to social ever more sophisticated understanding of the logic reality. that lies behind those techniques. Whereas qualitative research is sometimes un- It's much more difficult to teach qualitative dertaken for purely descriptive purposes-such as analysis as a series of rote procedures. In this case, the anthropologist's ethnography detailing ways of understanding must precede practice. In this chap- life in a previously unknown tribe-the rest of this ter, we begin 'with the links between research and chapter focuses primarily on the search for explan- theory in qualitative analysis. Then we examine atory patterns. As we'll see, sometimes the patterns some procedures that have proven useful in pursu- occur over time, and sometimes they take the form ing the theoretical aims. After considering some of causal relations among variables. Let's look at simple manual techniques, we'll take some com- some of the ways qualitative researchers uncover puter programs out for a spin. such patterns. linking Theory and Analysis Discovering Patterns As suggested in Chapter 10 and elsewhere in this John and Lyn Lofland (1995: 127-45) suggest six book, qualitative research methods involve a con- different ways of looking for patterns in a particular tinuing interplay between data collection and research topic Let's suppose you're interested in analyzing child abuse in a certain neighborhood. qualitative analysis The nonnumerical examina- Here are some questions you might ask yourself to tion and interpretation of observations, for the pur- make sense out of your data: pose of discovering underlying meanings and pat- terns of relationships. This is most typical of field 1. Frequwcies: How often does child abuse occur research and historical research. among families in the neighborhood under study? (Realize that there may be a difference between the frequency and what people are willing to tell you.)

Linking Theory and Analysis 379 2. Magnitudes: What are the levels of abuse? How Sometimes, though, it's useful to have even a par- brutal are they? tial explanation of overall orientations and actions. 3. STructures: What are the different types of abuse: You may also recall Chapter 1's introduction of physical. mental. sexual? Are they related in idiographic explanation, wherein we attempt to any particular manner? understand a particular case fully. In the voting ex- ample, we would attempt to learn everything we 4. Processes: Is there any order among the elements could about all the factors that came into play in of structure? Do abusers begin 'with mental determining one person's decision on how to vote. abuse and move on to physical and sexual This orientation lies at the base of what Huberman abuse, or does the order of elements vary? and .Miles call a case-oriented analysis. 5. Causes: What are the causes of child abuse? Is In a case-oriented analysis, we would look it more common in particular social classes or more closely into a particular case, say, Case among different religious or ettmic groups? 005, who is female, middle-class, has parents Does it occur more often during good times with high expectations, and so on. These are, or bad? however. \"thin\" measures. To do a genuine case analysis, we need to look at a full history 6. COllsequences: How does child abuse affect the of Case 005; Nynke van der Molen, whose victims, in both the short and the long term? mother trained as a social worker but is bitter What changes does it cause in the abusers? over the fact that she never worked outside the home, and whose father wants Nynke to work For the most part, in examining your data in the family florist shop. Chronology is also you'll look for patterns appearing across several ob- important: two years ago, Nynke's closest friend servations that typically represent different cases decided to go to college, just before Nynke be- under study. A. .Michael Huberman and Matthew gan work in a stable and just before Nynke's .Miles (1994: 435f) offer two strategies for cross- mother showed her a scrapbook from social case analysis: the variable-oriented and the case- work schooL Nynke then decided to enroll in oriented analysis. Variable-oriented analysis is veterinary studies. similar to a model we've already discussed from time to time in this book. If we were trying to pre- (1994.436) dict the decision to attend college, Huberman and Miles suggest. we might consider variables such as This abbreviated cOflilllentary should give you \"gender, socioeconomic status, parental expecta- some idea of the detail involved in this type of tions, school performance, peer support. and deci- analysis. Of course, an entire analysis would be sion to attend college\" (1994: 435). Thus, we would more extensive and pursue issues in greater depth. determine whether men or women were more This full. idiographic examination, however. tells us likely to attend college. The focus of our analysis nothing about people in general. It offers nothing would be on interrelations among variables, and in the way of a theory about why people choose to the people observed would be primarily the carriers attend college. of those variables. Even so, in addition to understanding one per- Variable-oriented analysis may remind you of son in great depth, the researcher sees the critical the discussion in Chapter 1 that introduced the idea of nomothetic explanation. The aim here is to variable-oriented analysis An analysis that de- achieve a partial. overall explanation using a rela- scribes andlor eX1llains a particular variable\" tively few number of variables. The political pollster case-oriented analysis An analysis that aims to who attempts to explain voting intentions on the understand a particular case or several cases by look- basis of two or three key variables is using this ap- ing closely at the details of each\" proach. There is no pretense that the researcher can predict every individual's behavior nor even explain anyone person's motivations in full.

380 Chapter 13: Qualitative Data Analysis elements of the subject's experiences as instances with observations rather than hypotheses and seeks of more general social concepts or variables. For to discover patterns and develop theories from the example, Nynke's mother's social work training can ground up, with no preconceptions, though some also be seen as \"mother's education.\" Her friend's research may build and elaborate on earlier decision can be seen as \"peer influence.\" More spe- grounded theories. cifically, these could be seen as independent vari- ables having an in1pact on the dependent variable Grounded theory was first developed by the of attending college. sociologists Barney Glaser and Anselm Strauss (1967) in an attempt to come to grips with their Of course, one case does not a theory make- clinical research in medical sociology. (You can hear hence Huberman and Miles refer to cross-case Glaser discuss grounded theory on the web at analysis, in which the researcher turns to other http://www.groundedtheory.com/vidseries 1.htmL) subjects, looking into the full details of their lives as Since then, it has evolved as a method, with the co- well but paying special note to the variables that founders taking it in slightly different directions. seemed important in the first case. How much and The following discussion will deal with the basic what kind of education did other subjects' mothers concepts and procedures of the Grounded The- have? Is there any evidence of close friends attend- ory Method (GTM). ing college? In addition to the fundamental, inductive tenet Some subsequent cases will closely parallel the of building theory from data, GTM employs the first one in the apparent impact of particular vari- constant comparative Inethod. As Glaser and ables. Other cases \\vill bear no resemblance to the Strauss originally described this method, it involved first. These latter cases may require the identifica- four stages (1967: 105-13): tion of other important variables, which may invite the researcher to explore why some cases seem to 1. \"Comparing incidents applicable to each cate- reflect one pattern while others reflect anotheL gory.\" As Glaser and Strauss researched the re- actions of nurses to the possible death of pa- Grounded Theory Method tients in their care, the researchers found that the nurses were assessing the \"social loss\" atten- The cross-case method just described should sound dant upon a patient's death. Once this concept somewhat familiaL In the discussion of grounded arose in the analysis of one case, they looked theory in Chapter 10, we saw how qualitative re- for evidence of the same phenomenon in other searchers sometimes attempt to establish theories cases. When they found the concept arising in on a purely inductive basis. This approach begins the cases of several nurses, they compared the different incidents. This process is similar to cross-case analysis An analysis that involves an conceptualization as described in Chapter 5- examination of more than one case; this can be ei- specifying the nature and dimensions of the ther a variable-oriented or case-oriented analysis. many concepts arising from the data. Grounded Theory Method (GTM) An inductive approach to research, introduced by Barney Glaser 2. \"Integrating categories and their properties.\" and Anselm Strauss, in which theories are generated Here the researcher begins to note relationships solely from an examination of data rather than be- among concepts. In the assessment of social ing derived deductively. loss, for example, Glaser and Strauss found that constant comparative method A component of nurses took special notice of a patient's age, ed- the Grounded Theory Method in which observa- ucation, and family responsibilities. For these tions are compared with one another and with the relationships to emerge, however, it was neces- evolving inductive theory. sary for the researchers to have noticed all these concepts. 3. \"Delimiting the theory.\" Eventually, as the pat- terns of relationships among concepts become

Unking'Theory and Analysis 381 clearer, the researcher can ignore some of the 1. Poinsettia a. Good luck concepts initially noted but evidently irrelevant 2. Horseshoe b. First prize to the inquiry. In addition to the number of cat- 3. Blue ribbon c. Christmas egories being reduced, the theory itself may be- 4. \"Say cheese\" d. Acting come simpleL In the examination of social loss, 5. \"Break a leg\" e. Smile for a picture for example, Glaser and Strauss found that the assessment processes could be generalized be- FIGURE 13-' yond nurses and dying patients: They seemed Matching Signs and their Meanings to apply to the ways all staff dealt with all pa- tients (dying or not). complexity. Morse code, etiquette, mathematics, music, and even highway signs are examples of 4. \"Writing theory.\" Finally, the researcher must semiotic systems.\" put his or her findings into words to be shared with others. As you may have already experi- There is no meaning inherent in any sign, enced for yourself, the act of communicating however. Meanings reside in minds. So, a particu- your understanding of something actually lar sign means something to a particular person. modifies and even improves your own grasp However, the agreements we have about the of the topic. In GTM, the 'I\"VTiting stage is re- meanings associated with particular signs make garded as a part of the research process. A later semiotics a social science. As Manning and Cullum- section of this chapter (on memoing) elaborates Swan point out: on this point. For example, a lily is an expression linked con- This brief overview should give you an idea of ventionally to death, Easter, and resurrection as how grounded theory proceeds. The many tech- a content. Smoke is linked to cigarettes and to niques associated with GTM can be found both cancer, and Marilyn Monroe to sex. Each of in print and on the web. One key publication is these connections is social and arbitrary, so that Anselm Strauss and Juliet Corbin's Basics of Qualita- many kinds of links exist between expression tive Research (1998), which elaborates on and ex- and content. tends many of the concepts and techniques found in the original Glaser/Strauss volume. On the web, (1994 466) you can search for \"grounded theory\" to see a wealth of articles. To explore this contention, see if you can link the signs with their meanings in Figure 13-1. I'm GTM is only one analytical approach to quali- confident enough that you know all the \"correct\" tative data. In the remainder of this section, we'll associations that there's no need for me to give the take a look at some other specialized techniques. answers. (OK, you should have said lc, 2a, 3b, 4e, 5d.) The point is this: What do any of these signs Semiotics have to do with their \"meanings\"? Draft an e-mail message to a Martian social scientist explaining the Semiotics is commonly defined as the \"science of logic at work here. (You might want to include signs\" and has to do with symbols and meanings. some \"emoticons\" like: ) -another example of It's commonly associated with content analysis, semiotics. ) which was discussed in Chapter 1L though it can be applied in a variety of research contexts. semiotics The study of signs and the meanings as- Peter Manning and Betsy Cullum-Swan (1994: sociated with them. This is commonly associated 466) offer some sense of the applicability of semi- otics, as follows: \"Although semiotics is based on with content analysis. . language, language is but one of the many sign sys- tems of varying degrees of unity, applicability, and

382 Chapter 13: Qualitative Data Analysis FIGURE 13-2 Mixed Signals? Although there is no doubt a story behind each pictures found in magazines and newspapers. The of the linkages in Figure 13 -1, the meanings you overt purpose of the ads, of course, was to sell spe- and I \"know\" today are socially constructed. Semi- cific products. But what else was communicated, otic analysis involves a search for the meanings in- Goffman asked. What in particular did the ads say tentionally or unintentionally attached to signs. about men and women? Consider the sign shown in Figure 13-2, from a Analyzing pictures containing both men and hotel lobby in Portland, Oregon. What's being com- women, Goffman was struck by the fact that men municated by the rather ambiguous sign? The first were almost always bigger and taller than the sentence seems to be saying that the hotel is up to women accompanying them. (In many cases, in date with the current move away from tobacco in fact, the picture managed to convey the distinct the United States. Guests who want a smoke-free impression that the women were merely accompa- environment need look no farther: This is a healthy nying the men.) Although the most obvious expla- place to stay. At the same time, says the second nation is that men are, on average, heavier and sentence, the hotel would not like to be seen as in- taller than women, Goffman suggested the pattern hospitable to smokers. There's room for everyone had a different meaning: that size and placement under this roof. No one need feel excluded. This implied status. Those larger and taller presumably sign is more easily understood within a marketing had higher social standing-more power and au- paradigm than one of logic. thority (1979: 28). Goffman suggested that the ads communicated that men were more important The \"signs\" examined in semiotics, of course, than women. are not limited to this kind of sign. Most are quite different, in fact Signs are any things that are as- In the spirit of Freud's comment that \"some- signed special meanings. They can include logos, times a cigar is just a cigar\" (he was a smoker), how animals, people, and consumer products. Some- would you decide whether the ads simply reflected times the symbolism is a bit subtle. A classic analy- the biological differences in the average sizes of sis can be found in Erving Goffman's Gender Adver- men and women or whether they sent a message tisemellts (1979). Goffman focused on advertising about social status? In part, Goffman's conclusion

Linking Theory and Analysis 383 was based on an analysis of the exceptional cases: of the way we converse with one another. In the those in which the women appeared taller than the examination of ethnomethodology in Chapter 10, men. In these cases, the men were typically of a you saw some examples of conversation analysis. lower social status-the chef beside the society ma- Here we'll look a little more deeply into that tron, for example. This confirmed Goffman's main technique. point that size and height indicated social status. David Silverman (1999), reviewing the work of The same conclusion was to be drawn from pic- other CA theorists and researchers, speaks of three Hires with men of different heights. Those of higher fundamental assumptions. First, conversation is a status were taller. whether it was the gentleman socially structured activity. Like other social struc- speaking to a waiter or the boss guiding the work of tures, it has established rules of behavior. For ex- his younger assistants. Where actual height was un- ample, we're expected to take turns, with only one clear, Goffman noted the placement of heads in the person speaking at a time. In telephone conversa- picture. The assistants were crouching dovvn while tions, the person answering the call is expected to the boss leaned over them. The servant's head was speak first (e.g., \"Hello\"). You can verify the exis- bowed so it was lower than that of the master. tence of this rule, incidentally, by picking up the phone without speaking. You may recall that this is The latent message conveyed by the ads, then, the sort of thing ethnomethodologists tend to do. was that the higher a person's head appeared in the ad, the more important that person was. And in the Second, Silverman points out that conversa- great majority of ads containing men and women, tions must be understood contextually. The same the former were clearly portrayed as more impor- utterance will have totally different meanings in tant. The subliminal message in the ads, whether different contexts. For example, notice how the intended or not, was that men are more powerful meaning of \"Same to you!\" varies if preceded by than women and enjoy a higher status. \"I don't like your looks\" or by \"Have a nice day.\" Goffman examined several differences besides Third, CA aims to understand the structure and physical size in the portrayal of men and women. meaning of conversation through excruciatingly As another example, men were typically portrayed accurate transcripts of conversations. Not only are in active roles, women in passive ones. The (male) the exact words recorded, but all the uhs, ers, bad doctor examined the child while the (female) nurse grammar, and pauses are also noted. Pauses, in or mother looked on, often admiringly. A man fact. are recorded to the nearest tenth of a second. guided a woman's tennis stroke (all the while keep- ing his head higher than hers). A man gripped the The practical uses of this type of analysis are reins of his galloping horse, while a woman rode be- many. Ann Marie Kinnell and Douglas Maynard hind him with her arms wrapped around his waist. (1996), for example, analyzed conversations be- A woman held the football, while a man kicked it tween staff and clients at an HN-testing clinic to A man took a photo, which contained only women. examine how information about safe sex was com- municated. An10ng other things, they found that Goffman suggested that such pictorial patterns the staff tended to provide standard information subtly perpetuated a host of gender stereotypes. rather than try to speak directly to a client's specific Even as people spoke publicly about gender equal- circumstances. Moreover. they seemed reluctant ity, these advertising photos established a quiet to give direct advise about safe sex, settling for backdrop of men and women in the \"proper roles.\" information alone. Conversation Analysis These discussions should give you some sense of the variety of qualitative analysis methods Etlmomethodology, as you'll recall, aims to un- cover the implicit assumptions and structures in so- conversation analysis (CA) A meticulous analysis cial life. Conversation analysis (CA) seeks to pur- of the details of conversation, based on a complete sue that aim through an extremely close scrutiny transcript that includes pauses, hems, and also haws.

384 Chapter 13: Qualitative Data Analysis available to researchers\" Now let's look at some of notes have been catalogued by topic, retrieving the data-processing and data-analysis techniques those you need should be straightforward. As a commonly used in qualitative research\" simple format for coding and retrievaL you might have created a set of file folders labeled with vari- Qualitative Data Processing ous topics, such as \"History.\" Data retrieval in this case means pulling out the \"History\" folder and Let me begin this section with a warning. The ac- rifling through the notes it contains until you find tivity we're about to examine is as much art as sci- what you need. ence\" At the very least there are no cut-and-dried steps that guarantee success\" As you'll see later in this chapter, there are now sophisticated computer programs that allow for a It's a lot like learning how to paint with water- faster, more certain, and more precise retrieval pro- colors or compose a symphony\" Education in such cess\" Rather than looking through a \"History\" file, activities is certainly possible, and university courses you can go directly to notes dealing with the \"Earli- are offered in both. Each has its own conventions est History\" or the \"Founding\" of the movement. and techniques as well as tips you may find useful as you set out to create art or music. However, in- Coding has another, even more important pur- struction can carry you only so far. The final prod- pose\" As discussed earlier, the aim of data analysis is uct must come from you\" Much the same can be the discovery of patterns among the data, patterns said of qualitative data processing. that point to theoretical understandings of social life\" The coding and relating of concepts is key to This section presents some ideas relating to the this process and requires a more refined system coding of qualitative data, writing memos, and than a set of manila folders. In this section, we'll as- mapping concepts graphically. Although far from a sume that you'll be doing your coding manually. \"how-to\" manual, these ideas give a useful starting The next section of the chapter will illustrate the use point for finding order in qualitative data. of computer programs for qualitative data analysis\" Coding Coding Units Whether you've engaged in participant observa- As you may recall from the earlier discussion of tion, in-depth interviewing, collecting biographical content analysis, for statistical analysis it's impor- narratives, doing content analysis, or some other tant to identify a standardized unit of analysis prior form of qualitative research, you'll now be in the to coding. If you were comparing American and possession of a growing mass of data-most typi- French novels, for example, you might evaluate cally in the form of textual materials. Now what do and code sentences, paragraphs, chapters, or whole you do? books. It would be important, however, to code the same units for each novel analyzed\" This unifor- The key process in the analysis of qualitative mity is necessary in a quantitative analysis, as it al- social research data is coding-classifying or catego- lows us to report something like \"23 percent of the rizing individual pieces of data-coupled with paragraphs contained metaphors.\" This is only pos- some kind of retrieval system\" Together, these pro- sible if we've coded the same unit-paragraphs- cedures allow you to retrieve materials you may in each of the novels\" later be interested in. Coding data for a qualitative analysis, however, Let's say you're chronicling the growth of a is quite different. The concept is the organizing prin- social movement\" You recall writing up some notes ciple for qualitative coding. Here the units of text about the details of the movement's earliest begin- appropriate for coding will vary within a given doc- nings\" Now you need that information. If all your ument. Thus, in a study of organizations, \"Size\" might require only a few words per coding unit, whereas \"Mission\" might take a few pages\" Or, a

Qualitative Data ProcesSing 385 lengthy description of a heated stockholders meet- that have been generated by prior theory\" In that ing might be coded as \"Internal Dissent.\" case, then, the codes would be suggested by the theory, in the form of variables. Realize also that a given code category may be applied to text materials of quite different lengths\" In this section, however, we're going to focus For example, some references to the organization's on the more common processes of open coding, axial mission may be brief, others lengthy. Whereas stan- coding and selective coding. Strauss and Corbin (1998: dardization is a key principle in quantitative analy- 102) describe open coding as follows: sis, this is not the case in qualitative analysis. To uncover, name, and develop concepts, we Coding as aPhysical Act must open up the text and expose the thoughts, ideas, and meanings contained therein\" With- Before continuing 'with the logic of coding, let's out this first analytic step, the rest of the analy- take a moment to see what it actually looks like\" sis and the conununication that follows could John and Lyn Lofland (1995: 188) offer this de- not occur. Broadly speaking, during open cod- scription of manual filing: ing, data are broken down into discrete parts, closely examined, and compared for similarities Prior to the widespread availability of personal and differences\" Events, happenings, objects, computers beginning in the late 1980s, coding and actions/interactions that are found to be frequently took the specific physical form of conceptually similar in nature or related in filing\" The researcher established an expanding meaning are grouped under more abstract con- set of file folders with code names on the tabs cepts termed \"categories.\" and physically placed either the item of data it- self or a note that located it in the appropriate Although the analysiS of data will quickly ad- file folder. \"\". Before photocopying was easily vance to an iterative interplay of the three types of available and cheap, some fieldworkers typed coding, open coding is the logical starting point. their fieldnotes with carbon paper, wrote codes Beginning with some body of text (part of an inter- in the margins of the copies of the notes, and view, for example), you read and reread a passage, cut them up v'lith scissors\" They then placed seeking to identify the key concepts contained the resulting slips of paper in corresponding within it. Any particular piece of data may be given file folders. several codes, reflecting as many concepts. For ex- ample, notice all the concepts contained in this As the Loflands point out personal computers comment by a student interviewee: have greatly simplified this task\" However, the im- age of slips of paper that contain text and are put in I thought the professor should have given folders representing code categories is useful for un- me at least partial credit for the homework derstanding the process of coding. In the next sec- I turned in. tion, when I suggest that we code a textual passage with a certain code, imagine that we have the pas- Some obvious codes are \"Professor,\" \"Homework,\" sage typed on a slip of paper and that we place it in and \"Grading.\" The result of open coding is the a file folder bearing the name of the code\" When- identification of numerous concepts relevant to the ever we assign two codes to a passage, imagine plac- subject under study\" The open coding of more and ing duplicate copies of the passage in two different more text vvill lengthen the list of codes. folders representing the two codes\" Creating Codes open coding The initial classification and labeling of concepts in qualitative data analysis. In open cod- So, what should your code categories be? Glaser ing, the codes are suggested by the researchers' ex- and Strauss (1967: IOlf) allow for the possibility of amination and questioning of the data. coding data for the purpose of testing hypotheses

386 Chapter 13: Qualitative Data Analysis Axial coding aims to identify the core concepts 20:13 If a man lies vvith a male as with a woman, in the study. Although axial coding uses the results both of them have conmlitted an abomina- of open coding, more concepts can be identified tion; they shall be put to death, their blood through continued open coding after the axial cod- is upon them. ing has begun. Axial coding involves a regrouping of the data, in which the researcher uses the open Although the point of view expressed here code categories and looks for more analytical con- seems unambiguous, you might decide to examine cepts. For example, the passage just given also car- it in more depth. Perhaps a qualitative analysis of ries the concept of \"perceptions of fairness,\" which Leviticus can yield a fuller understanding of where might appear frequently in the student interviews, these injunctions against homosexuality fit into the thereby suggesting that it's an important element in larger context of Judea-Christian morality. understanding students' concerns. Another axial code reflected in the student comment might be Let's start our analysis by exanlining the two \"power relationships,\" because the professor is seen passages just quoted. We might begin by coding to exercise power over the student. each passage with the label \"Homosexuality.\" This is clearly a key concept in our analysis. Whenever we Selective coding seeks to identify the central focus on the issue of homosexuality in our analysis code in the study: the one that the other codes all of Leviticus, we want to consider these two passages. related to. Both of the axial codes just mentioned might be restructured as aspects of a more general Because homosexuality is such a key concept, concept: \"professor-student relationships.\" Of let's look more closely into what it means within course, in a real data analysis, decisions such as the data under study. We first notice the way homo- the ones we've been discussing would arise from sexuality is identified: a man lying with a man \"as masses of textual data, not from a single quotation. \\'\\rith a woman.\" Although we can imagine a lawyer The basic notion of the Grounded Theory Method seeking admission to heaven saying, \"But here's my is that patterns of relationships can be teased out of point; if we didn't actually lie down ...\" it seems an extensive, in-depth examination of a large body safe to assume the passage refers to having sex, of observations. though what specific acts might or might not be included isn't clear. Here's a concrete example to illustrate how you might engage in this form of analysis. Suppose Notice, however, that the injunctions appear to you're interested in the religious bases for homo- concern male homosexuality only; lesbianism is not phobia. You've interviewed some people opposed mentioned. In our analysis, then, each of these pas- to homosexuality who cite a religious basis for their sages might also be coded \"Male Homosexuality.\" feelings. Specifically, they refer you to these pas- Tilis illustrates two more aspects of coding: (1) Each sages in the Book of Leviticus (Revised Standard unit can have more than one code and (2) hierar- Version): chical codes (one induded witllin another) can be used. Now each passage has two codes assigned to it. 18:22 You shall not lie with a male as with a woman; it is an abomination. An even more general code might be intro- duced at this point: \"Prohibited Behavior:' This is axial coding A reanalysis of the results of open important for two reasons. First, homosexuality is coding in Grounded Theory Method, aimed at iden- not inherently wrong, from an analytical stand- tifying the important, general concepts. point. The purpose of the study is to examine the way it's made wrong by the religious texts in ques- selective coding In Grounded Method Theory. this tion. Second, our study of Leviticus may turn up analysis builds on the results of open coding and ax- other beha\\riors that are prohibited. ial coding to identify the central concept that orga- nizes the other concepts that have been identified in There are at least two more critical concepts in a body of textual materials. the passages: \"Abomination\" and \"Put to Death:' Notice that although these are clearly related to \"Prollibited Behavior,\" they are hardly the same.

Qualitative Data Processing 387 Parking without putting money in the meter is pro- 11.:16 the ostrich, the nighthawk, the sea gull, hibited, but few would call it an abomination and the hawk according to its kind, fewer still would demand the death penalty for that transgression. Let's assign these two new codes to 1I :17 the owl, the cormorant, the ibis, our first two passages. 11.18 the water hen, the pelican, the carrion At this point. we want to branch out from the vulture, two key passages and examine the rest of Leviticus. We therefore examine and code each of the re- 11:19 the stork, the heron according to its kind, maining chapters and verses. In our subsequent the hoopoe, and the bat. analyses, we'll use the codes we have already and add new ones as appropriate. When we do add 11020 All winged insects that go upon all new codes, it 'will be important to review the pas- fours are an abomination to you. sages already coded to see whether the new codes apply to any of them. 11.:41 Every swarming thing that swarms upon Here are the passages we decide to code the earth is an abomination; it shall not be \"Abomination.\" (I've boldfaced the abOminations.) eaten. 11:42 Whatever goes on its belly, and whatever goes on all fours, or whatever has many 7:18 If any of the flesh of the sacrifice of his feet, all the swarming things that swarm peace offering is eaten on the third upon the earth, you shall not eat; for they day, he who offers it shall not be accepted, are an abomination. neither shall it be credited to him; it shall be an abomination, and he who eats of it 11:43 You shall not make yourselves abominable shall bear his iniquity. with any swarming thing that swarms; and you shall not defile yourselves with them, lest you become unclean. 7:21 And if anyone touches an unclean 18022 You shall not lie with a male as with a thing, whether the uncleanness of man or woman; it is an abomination. an unclean beast or any unclean abomina- tion, and then eats of the flesh of the 19. 6 It shall be eaten the same day you offer it, or sacrifice of the LORD's peace offerings, on the morrow; and anything left over until that person shall be cut off from his people. the third day shall be burned \"rith fire. 11:10 But anything in the seas or the rivers 19:7 If it is eaten at all on the third day, it is that has not fins and scales, of the an abomination; it will not be accepted, swarming creatures in the waters and of the living creatures that are in the waters, 19.8 and everyone who eats it shall bear his is an abomination to you. iniquity, because he has profaned a holy thing of the LORD; and that person shall be 11:1 I They shall remain an abomination to cut off from his people. you; of their flesh you shall not eat, and their carcasses you shall have in 20: 13 If a man lies with a male as with a abomination. woman, both of them have committed an abomination; they shall be put to death, 1112 Everything in the waters that has not their blood is upon them. fins and scales is an abomination to you. 20:25 You shall therefore make a distinction be- tween the clean beast and the unclean, and 11:13 And these you shall have in abomination between the unclean bird and the clean; among the birds, they shall not be eaten, you shall not make yourselves abom- they are an abomination: the eagle, the inable by beast or by bird or by any- vulture, the osprey, thing with which the ground teems, which I have set apart for you to hold 1I: 14 the kite, the falcon according to its kind, unclean . l1'I5 every raven according to its kind,

388 Chapter 13: Qualitative Data Analysis Male homosexuality, then, isn't the only abomi- blocks in front of blind people. In chapter 19, verse nation identified in Leviticus. As you compare these 19, Leviticus quotes God as ordering, \"You shall not passages, looking for similarities and differences, it let your cattle breed with a different kind; you shall vvill become apparent that most of the abominations not sow your field with two kinds of seed; nor shall have to do with dietary rules-specifically those there come upon you a garment of cloth made of potential foods deemed \"unclean.\" Other abomina- two kinds of stuff.\" Shortly thereafter, he adds, tions flow from the mishandling of ritual sacrifices. \"You shall not eat any flesh with the blood in it. \"Dietary Rules\" and \"Ritual Sacrifices\" thus repre- You shall not practice augury or witchcraft. You sent additional codes to be used in our analysis. shall not round off the hair on your temples or mar the edges of your beard.\" Tattoos were prohibited, Earlier, I mentioned the death penalty as an- though Leviticus is silent on body piercing. Refer- other concept to be explored in our analysis. When ences to all of these practices would be coded \"Pro- we take this avenue, we discover that many behav- hibited Acts\" and perhaps given additional codes as iors besides male homosexuality warrant the death well (recall \"Dietary Rules\"). penalty. Among them are these: I hope this brief glimpse into a possible analysis 20.:2 Giving your children to Molech (human vvill give you some idea of the process by which sacrifice) codes are generated and applied. You should also 20.:9 Cursing your father or mother have begun to see how such coding would allow 20.:10 Adultery with your neighbor's wife you to better understand the messages being put 20.:11 Adultery with your father's wife forward in a text and to retrieve data appropriately 20.:12 Adultery with your daughter-in-law as you need them. 20.:14 Taking a wife and her mother also 20.:15 Men having sex with animals (the animals Memoing are to be killed, also) 20.:16 Women having sex with animals In the Grounded Theory Method, the coding pro- 20.:27 Being a medium or vvizard cess involves more than simply categorizing chunks 24:16 Blaspheming the name of the Lord of text. As you code data, you should also be using 2417 Kminga man the technique of memoing-writing memos or notes to yourself and others involved in the project. As you can see, the death penalty is broadly Some of what you write during analysis may end applied in Levicitus: everything from swearing to up in your final report; much of it will at least stim- murder, including male homosexuality somewhere ulate what you write. in between. In GTM, these memos have a special signifi- An extended analysis of prohibited behavior, cance. Strauss and Corbin (1998: 217) distinguish short of abomination and death, also turns up a three kinds of memos: code notes, theoretical notes, lengthy list. Among them are slander, vengeance, and operational notes. grudges, cursing the deaL and putting stumbling Code notes identify the code labels and their memoing Writing memos that become part of the meanings. This is particularly important because, as data for analysis in qualitative research such as in all social science research, most of the terms we grounded theory. Memos can describe and define use with technical meanings also have meanings in concepts, deal with methodological issues, or offer everyday language. It's essential, therefore, to write initial theoretical formulations. down a clear account of what you mean by the codes used in your analysis. In the Leviticus analy- sis, for example, you would want a code note re- garding the meaning of \"Abomination\" and how you've used that code in your analysis of text.

Qualitative Data Processing 389 Theoretical notes cover a variety of topics: elemental memos. A given project may see the CTe- reflections of the dimensions and deeper meanings ation of several sorting memos dealing with differ- of concepts, relationships among concepts, theoret- ent aspects of the project. ical propositions, and so on. All of us have rumi- nated over the nature of something, trying to think Finally, the integrating memo ties together the it out, to make sense out of it. In qualitative data several sorting memos to bring order to the whole analysis, it's vital to write down these thoughts, project. It tells a coherent and comprehensive story; even those you'll later discard as useless. They casting it in a theoretical context. In any real proj- will vary greatly in length, though you should ect, however, there are many different ways of limit them to a single main thought so that you bringing about this kind of closure. Hence, the data can sort and organize them later. In the Leviticus analysis may result in several integrating memos. analysis, one theoretical note might discuss the way that most of the injunctions implicitly address Notice that whereas we often think of writing the behavior of men vvith women being mostly as a linear process, starting at the beginning and incidental. moving through to the conclusion, memoing is very different. It might be characterized as a pro- Operational notes deal primarily with method- cess of creating chaos and then finding order ological issues. Some will draw attention to data- within it. collection circumstances that may be relevant to understanding the data later on. Others will consist To explore this process further, refer to the of notes directing future data collection. works cited in this discussion and at the end of the chapter. You'll also find a good deal of infor- Writing these memos occurs throughout the mation on the web. For Barney Glaser's rules on data-collection and analysis process. Thoughts de- memoing, for example, you might go to http:// manding memos will come to you as you reread www.vlsm.org/gnm/gnm-gtm3.htmL Ultimately, notes or transcripts, code chunks of text, or discuss the best education in this process comes from prac- the project with others. It's a good idea to get in the tice. Even if you don't have a research project un- habit of writing out your memos as soon as possible derway, you can practice now on class notes. Or after the thoughts corne to you. start a journal and code it. John and Lyn Lofland (1995: 193f) speak of concept Mapping memoing somewhat differently, describing memos that corne closer to the final writing stage. The ele- It should be clear by now that qualitative data ana- mental memo is lysts spend a lot of time committing thoughts to pa- per (or to a computer file), but this process is not a detailed analytic rendering of some relatively limited to text alone. Often, we can think out rela- specific matter. Depending on the scale of the tionships among concepts more clearly by putting project, the worker may write from one to the concepts in a graphical format, a process called several dozen or more of these. Built out of concept mapping. Some researchers put all their selective codes and codings, these are the major concepts on a single sheet of paper, whereas most basic prose cannon fodder, as it were, of others spread their thoughts across several sheets of the project. paper, blackboards, magnetic boards, computer pages, or other media. Figure 13-3 shows how we (1995: 194) might think out some of the concepts of Goffman's The sorting memo is based on several elemental concept mapping The graphical display of con- memos and presents key themes in the analysis. cepts and their interrelations, useful in the formula- Whereas we create elemental memos as they come tion of theory. to mind, with no particular rhyme nor reason, we write sorting memos as an attempt to discover or create reason among the data being analyzed. A sorting memo will bring together a set of related

390 Chapter 13: Qualitative Data Analysis Active I statistics. The inlportance of the computer for qual- passive itative research has been somewhat more slowly appreciated. Some qualitative researchers were roles quick to adapt the basic capacities of computers to nonnumerical tasks, but it took a bit longer for FIGURE 13-3 programmers to address the needs of qualitative An Example of Concept Mapping research per se. Today, however, several powerful programs are available. examination of gender and advertising. (This im- age was created through the use of Inspiration, a Let's start this section with a brief overview of concept-mapping computer program.) some of the ways you can use basic computer tools in qualitative research. Perhaps only those who can Incidentally, many of the topics discussed in recall hours spent with carbon paper and White- this section have useful applications in quantitative Out can fully appreciate the glory of computers in as well as qualitative analyses. Certainly, concept this regard. \"Easier editing\" and \"easier duplication\" mapping is appropriate in both types of analysis. sin1ply do not capture the scope of the advance. The several types of memos would also be useful in both. And the discussion of coding readily applies Moving beyond the basic recording and storage to the coding of open-ended questionnaire re- of data, simple word-processing programs can be sponses for the purpose of quantification and statis- used for some data analysis. The \"find\" or \"search\" tical analysis. (We'll look at coding again in the command will take you to passages containing key- next chapter, on quantifying data.) words. Or, going one step further, you can type code words alongside passages in your notes so that Having noted the overlap of qualitative and you can search for those keywords later. quantitative techniques, it seems fitting now to ad- dress an instrument that is primarily associated Database and spreadsheet programs can also be with quantitative research but that is proving quite used for processing and analyzing qualitative data. valuable for qualitative analysts as well-the per- Figure 13-4 is a simple illustration of how some of sonal computer. the verses from Leviticus might be manipulated \"within a spreadsheet. The three columns to the left computer Programs represent three of the concepts we've discussed. An for Qualitative Data \"X\" means that the passage to the right contains that concept. As shown, the passages are sorted in The advent of computers-mainframe and per- such a way as to gather all those dealing with pun- sonal-has been a boon to quantitative research, ishment by death. Another simple \"sort\" command allowing the rapid calculation of extremely complex would gather all those dealing with sex, v'<lith ho- mosexuality, or any of the other concepts coded. This brief illustration should give you some idea of the possibilities for using readily available pro- grams as tools in qualitative data analysis. Happily, there are now a large number of programs created specifically for that purpose. Here's an excellent list prepared by sociologists at the University of Surrey, England (http://vvww.socsurrey.acuk/sru/ SRUl.htmI): The Ethnograph HyperQual HyperResearch HyperSoft

Computer Programs for Qualitative Data 391 xx x 20:13 If a man lies with a male as with a woman, both of them have committed an x x abomination; they shall be put to death, their blood is upon them. xx x 20:12 If a man lies with his daughter-in-law, both of them shall be put to death; they have committed incest, their blood is upon them. x 20:15 If a man lies with a beast, he shall be put to death; and you shall kill the beast. x 20:09 For every one who curses his father or his mother shall be put to death; he has cursed his father or his mother, his blood is upon him. x 20:02 Any man of the people of Israel, or of the strangers that sojourn in Israel, who gives any of his children to Molech shall be put to death. 18:22 You shall not lie with a male as with a woman; it is an abomination. FIGURE 13-4 Using aSpreadsheet for Qualitative Analysis NUD*IST materials already in existence-such as field notes Qualrus or, in this case, the verses of Leviticus-are im- QUALOG ported into the program. Menu-based commands Textbase Alpha do this easily, though the text must be in a plaintext SONAR format (Le., without word-processing or other Atlas,.ti formatting). This website also provides a brief description of Figure 13-5 shows how the text is displayed each of the programs listed, along vvith the price within 'l\\TUD*IST. For the illustrations in this sec- and contact, where available. tion, I've used the Macintosh version of NUD*IST. I'll use the Windows version in the film director il- Leviticus as Seen through NUD*/ST lustration, so you can note the difference and simi- larities in the two platforms. Let's take a closer look at how qualitative data analysis programs operate, by considering one To see the document, select its name in me of the programs just mentioned, NUD*IST (Non- \"Document Explorer\" window and click \"Browse.\" numeric Unstructured Data, Index Searching, and The text window can be resized and moved around Theorizing). Although each of the programs has the screen to suit your taste. somewhat different features and different ap- proaches to analysis, NUD*IST is one of the most Note the set of buttons in the upper right cor- popular programs, and it offers a fairly representa- ner of the illustration. These allow you to select tive view of the genre. We'll begin with a brief ex- portions of the text for purposes of editing, coding, amination of Leviticus, and then we'll examine a and other operations. project focused on understanding the experiences of women film directors. Now let's create a concept code: \"homosex.\" This will stand for references to male homosexual- Although it is possible to type directly into ity. Figure 13-6 shmvs what the creation of a con- NUD*IST the text materials to be coded, usually cept code looks like. As we create codes for our concepts, we can use them to code the text materials. Figure 13-7 illustrates how this is done. In the document browser, you can see that verse 20: 13 has been

392 Chapter 13: Qualitative Data Analysis Text Searches [0 J IRoot ofyourlrulexTree lrulexSean:lres [[I J Coding Status: Documenl Annotations IN/A Nrule Clil'board Definition: Exlended clil:kto crealJo more nodes. o Document BrowseI': 'leviticus': 3 - 3 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ~++ Oti-L I tiE DOCUMEtlT: Lev i t i cus *' No Header ++++++++++++++++++++++++++++-++++++++++++++++++++++++++++++++++++++.;.+++++++++++++ [Levi ticus : 1 - 425 ] Levi tic:us 1 Leviticus Online Docume The LORD col led Moses, and spoke to him from the tent of meetmg, saying, Levi1il:us \"Speak to the people of Isroel, and soy to them, When any man of you br i ngs on offer j ng to the LORD, you sha II br i n9 your Headel': offerinq of cottle from the herd or from the flock. \"If his-offering is 0 burnt offering from the herd, he sholl * No Header offer 0 mole wi thout blemish; he shall offer it at the door of the tent of meeting, that he may be accepted before the LORD; 4 he shall lay his hand upon the head of the burnt offering, and it shall be accepted for him to make atonement for him. 5 Then he sha lt kill the bu 11 before the LORD; and Aaron's sons the priests shall present the blood, and throw the blood t~ound about against the attar thot is at the door of the tent of meet i ng. And he shall flay the: burnt offering and cut it into pieces; and the sons of Aaron the priest shall put fire on the altar, and 1ay wood in order upon the fire; 8 and Aaron's sons the priests shall lay the pieces, the head, and the fat, in order upon the wood that is on the fire upon the a 1tar; but its entre i 1s and its legs he sha 11 wash with water. And FIGURE 13·5 How Text Materials Are Displayed in NUD'IST selected (indicated by the box outline around As text materials are coded, the program can this verse)\" Having done that. we click the button then be used for purposes of analysis\" As a simple labeled \"Add Coding\" (not shown in this illustra- example, we might want to pull together all the tion)\" This prompts the computer to ask us to iden- passages coded \"homosex.\" This would allow us tify the appropriate code\" The easiest way to re- to see them all at once, looking for similarities and spond is to click the \"Browse\" button, which differences\" presents you with a list of the current codes. In this example, I selected \"homosex\" and entered the Figure 13-8 shows how NUD*IST would bring code ill (100). together the passages referring to male homosexu- ality\" To do this, all you do is select the code name

computer Programs for Qualitative Data 393 ~'at [ 1J NOlle EXjJlorer TextSearches [0] Node: (100) IndexSearches [0] fmosex Doc1ll!lent Almofilfions Coding Status: Node ClipbOard !No coding Definition: . Node Information ... ~) I homosexl Definition: lpassage refers to male homosexualit}\\ I Created: ~ 9:40 am, Jan 4, 2000. I I II J I SMake ReJJ(Jrt! =MN'_e,mo _Bt:~Y's_e last Modified: 9:40 am, Jan 4, 2000. Document Explorer IIMake Report I Memo 1 ~ I Cancel] [1 0K IJ leviticus Online Document: ILevi1il:US Header: *NoHei!der L-IDt.ak(! ~ep0r!\"j Memo JL~~!owse Ilprope!:leill J jClose FIGURE 13·6 == Creating the Code \"homosex\" in the \"Node Explorer\" window and dick the Using NVivo to Understand Women \"Make Report\" button\" Film Directors, by Sandrine Zerbib This simple example illustrates the possibilities Sandrine Zerbib is a French sociologist interested opened up by a program designed specifically in understanding the special difficulties faced by for qualitative data analysis\" To get a little deeper women breaking into the male-dominated world of into its use, let's shift to a different research film direction\" To address this issue, she interviewed example\"

~------------~~No~d~e -Ex-plu-n'-Cl------------,---------------\"--- Free Nodes [0 J Node: (l') Index Tree Roo! [ I ] ITex!searchNOdeS L 100 hamasex In.deXSearches [0] Coding Status: DocUIllllnt Anno1ations Node Clipboani IN/A IMil\"!;; !!ep{frd [ Definition: leviticus IReSll1!S of tex! searches are placed here I Dowment Brows •'levltiws : 1H9 - 1751 cut him off from among his people. 7 Consecrate your'selves therefore, and be holy; for I am the LORD your God. 8 Keep my statutes, and do them; I am the LORD who sanctify you. 9 For everyone 1who curses his - = =C=od-e-s-e-le-c-t-ed- -te-x-t-a-t-n-o-d-e-...-=- ; 1 1 bpluotodto ids euapthon; hheimh. as clJrsed 113 IIli CI man commits adultery w both the adu l terer and the a Type a node address 11 The man who lies VJ i th his fa father's nakedness; both of 1(100) [ Select... blood is upon them. 12 If a man lies with his dough put to death; they have comm 0 Free Node them. 13 I f a mCln lies UJ i th a rna La as l_l_c_a_n_c__e_I_1 [,I~===O=K===I] cbolmomodi ttiesduapnonabthoemmi.nation; th 14 tIhfeya smhanalltakbeesbaurnweidfewainthd hfie~e', ID~ot~~e~~an~d ~t ~e~y,~~t l~Ot~t~~er~e ~........~1 may be no wickedness among you. 15 If a man lies with a beast, he shall be put to death; and you shall ki II the beast. 16 I f a woman approaches any beast and l ie5 with it, you sha II ki II the woman and the beast; they shall be put to death, their blood is upon them. 17 \"If a man takes his sister, a daughter of his father or a daughter of his mother, and sees her nakedness, and she fiGURE 13-7 Coding aText Passage

/rree Nodes [0 ] Node E>:plorer I Index Tree Root [ 1 ] Node: (100) 11l1Oll1DSex Tex! Searches [ 1 ] Index Searches [0 ] Coding Status: Document Annotations Node Clipboard 11 document 5 text-units_ Definition: passage refers to male homosexuality 0 Report on (toO) E!J8 .... ~ Q.S.R. NUO* I ST Power vers ion, revision 4.0. ~ Licensee: Earl Babb ie. PROJECT: Levi ticus, User Earl Bobbie, 6:138 pm, Jan 4, 20130. **(*1***0*0O*)*e*f*in**i *ti*o*n**: **********Ih*o*m**o*s*e*x************************************************ !~ passage re f ers to male homosexua l i ty [ ++++++++++++++++++++++++++++++++++++++++ I;; +++ ON-L INE DOCUMENT: Levi tiCL~S rc\" +++ Retr i eva l for th i s documen t : 5 uni ts out of 25313, = 13.213% ++ Text units 1591-1592: R; IMake.R~po.rt~ l._.f\\1.~m.o 22 'Iou shall not lie wi th a male as with a uJOman; it is an 1591 ~ 1592 abomination. I~ ++ Text uni ts 1749-1751 : I 13 I f a man lies wi th a male as wi th a woman, both of them hove 1749 commi tted an abomination; they shall be put to death, Leviticus their 17513 blood is upon them. 1751 0 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ It +++ Total number of text un its retr i eved = 5 +++. Retr i eva ls in 1 out of 1 documents, = 108%. +++ The documents OJ i th retr i eva ls have a total H so te~<t uni ts retrieved in these documen ts of 2530 text uni ts, - = 0.20%. +++ All documents have a total of 25313 text un its, \" so text un its found in these documen ts = {3. 213%. [ \"'- ,..++-+++++++++++++++++++++++++++++++++++++++++++++.++++++++++++++++++++++++++.+++++++ ~ hi!!.! _.-:.c c.' ... '~... '>'.'~ .••.••. ·'·1-;;/ .. Browse ! LI'!.c;perti;E!!J C!OS~~.J I L. .J IJh!iI!el!ep~'1:j Memo FIGURE 13-8 Reporting on \"homosex\"

396 Chapter 13: Qualitative Data Analysis 30 women directors in depth. Having compiled M\".iSi\",;g.li:!m H§RM _lo'xJ hours of recorded interviews, she turned to ++++++++++++.++++.++++++++++++++++++++++++++++++++++++++~J NUD*IST as a vehicle for analysis. Let's have her de- +++ ON-LINE DOCUMENT: Joy scribe the ongoing process in her own words. , No Header Most software for qualitative analysis allows re- +++++++++++++++++++++++++++++++++++++++++++++++++++++.+- searchers to simultaneously analyze several in- terviews from different interviewers. However. [Joy 1 - 113 J I find it more efficient to start by importing Telephone InteN18V! With Joy only one interview into NUD*IST. Because you will have transcribed or at least read your inter- Researcher OK So could you tell me about your experience, \\vhat's views beforehand, you may be able to select the interview you think vvill be most fruitful. unique about what you do? You should trust yourself, because you are be- Joy OK. Certainly the beginnings of dlrectmg ViBre very unique. I have coming an expert in what you are currently studying and also because comparing and con- been doing this for about. 16 years yeah trasting interviews should help you get a sense ,A.nyway. at that time. 16 years ago. given today, there were very few of how accurate your analysis is. women directing. I mean very fev/. I think maybe a handful. After having completed about 30 inter- .A.nd [1 was somethtng that I really wanted to do. 1had been other jobs views with women filmmakers, I had a sense of In the business and was successful at them but directing was directing what the main themes were, because they kept was something I really really really felt a calling. coming up in each interview. Nevertheless, I .\"nd I went to school, and I studied not at a famous USC or UCLA school. needed a tool for synthesizing those pages and But In a less reno\"1med, then less renovvned, and more reno\\>'med now pages of interviews. I chose to start vvith my in- And I went off and made a couple of films. Small films 10 minutes terview with Joy. I had made a note to myself films. And 3D minutes films to use her interview as a starting point. An Gat people together. to contacts that I had in my other profession ., and older film director. she seemed to have strong points she wanted to get across. ask them to do those films for the joy of just doing a film Its certainly very prevalent today when people need to do something, In Figure 13-9, my interview vvith Joy has usually, people want crews, and so and so forth, with the experience, been imported as a \"text only\" file. (Only part So I went off and did these two films withoul really any ktnd of ngorous of the file is visible in the window.) training of like the USC or UClP;s graduates In your own coding, remember that In other words 1never had to do a film as part of a graduate course, or NUD*IST only reads text documents. There is no need to get fancy with your interview tran- anything like that scription; all formats are erased. At this point They did not have that sort oithing. you are ready to enjoy the coding process. You It was just a matter of gOtng out and making a film and try to see what can simply highlight words, sentences, or sec- tions and add nodes (Le., codes) to it. The first you could come up with step is to create \"free nodes,\" that is, nodes in- .A.nd try to see if you did not stumble too much, and so on and so forth, dependent of one another. How much text you should highlight per code is a decision you will So ¥filh those exercises, the two that I made and the one in particular, have to make. However. keep in mind that you which WdS a little more. had more production, \\vill have to use those quotes in the writing I went aut to pursue a directing career, part of your research. You will need to be con- Kno\\ving full well how drnicult It was going to be, vincing. You also want to deconstruct the In as much as people recognized me for other disciples and not for whole interview. Try to not leave anything out. It is easier to forgo using a quote because you directing <I FIGURE 13-9 Text of Interview with '~oy\" have found a better one later than to have nothing to use because you were not consistent enough in your dissection of the interview. When you create a node, you first want to use wide categories that would be more inclu- sive of other potential quotes. But you also want to be specific enough for your coding sys- tem to have validity. In Figure 13-10, for in- stance, I have created the free node \"past\" be- cause my interviewee referred to the past as being extremely challenging for women who wanted to be film directorso There were very few women directors back then, many fewer than today I decided to add a definition of this node so that I could remember why I llsed \"past\" as a node. I also anticipated having an- other free node called \"today.\" Then I could move the \"change\" node to the index tree root

Computer Programs for Qualitative Data 397 Node: (F 13) were very dear and veri sweat and sort of drew a line flFrea Nodes/pas! you know. kind of demanded that, you know, I come up to Coding St.tus: because it was either that or die, you know 0augh) 11 document,5 text-units< being into this mist. Definition: that and that was a big success, and as a result of that cess. I was asked to do anolher one and another and another Node Infonn.tion... IlFree Nodes! Definition: perception of the past in terms of gender inequalityl Created: 4:24 am, Dec 28, 1999. last Modified: 4:24 am, Dec 28, 1999. Make Report I Memo Browse I. Make Reportl about that anymore mdependent CIrCUS ! creating many small films r ; - - - - - - - - - - -Online Document:: \\41th women directors, and I think women directors needed an outlet. you knOVl. an outlet, s-ince they werenl getting the kind of work that they OOY should have gotten, or should gel Header. And 1\\\"5 that way 10 this day 1:--:-:[<-0\"'He-a--:do-_'-------- Of course there are more women directing today, than they were when I started but there are more shows, and more network, and more cable, and more everjthmg So the ineq'Jallties still exist In term of the ratio of work 'iersus hWI many women work. ,And I donl know frankly, you know. ho''cY to solve It 1have talked to the preSident of the Directors Guild who IS most sympathetic, and he has sent out, you know. letters to productions heads and studios. and so on and so forth, saYlOg \"please give these women a chance. they are talented, some oflhem. you knov\", a lot ofCfE-dit under FIGURE 13-10 Creating the Code \"past\" and create \"past\" and \"today\" as subnodes un- under each node. You can attach a different der \"change<\" node to a quote you have wrongly coded. It is preferable to start with free nodes before you In Figure 13-11, I have highlighted a pas- build a hierarchy of codes (or tree), because it sage that deals \\vith several things. Joy talks takes time and patience to understand how cat- about the Directors Guild of America (DGA. or egories are linked to one another. Coding other the directors' union) and more specifically interviews should help you organize your about the efforts of its president.. She also ex- coding system. presses her feelings toward gender inequalitieso According to her. having talent is not enough Figure 13-13 illustrates my decision to im- in Hollywood if there is a bias against women. I port two more interviews, Berta's and Queena's. decided to add two nodes to this quote, \"DGA,\" I could browse all three interviews on the same which I needed to create, and \"discrimination,\" screen. Because it was still early in the analysis which I had already created. process, I chose to analyze these two new inter- views one by one. It was now starting to make In Figure 13-12, I have attempted to trans- sense; I was starting to see patterns. NUD*IST form some free nodes into index trees. The let me keep records of number of occurrences software is flexible enough for me to move each node was attached to a quote, not only nodes, rename them, or see what quotes are

398 Chapter 13: Qualitative Data Analysis Node: (F 13) iFree Nodesfpast It was no question, being mto this mist r Coding Status: So. So that was that and that was a big success. and as a result oftha! r:-:---..,.-;:-:--,-.,,---- show. and success, I was asked to do another one and another and another }1 document, 5 text-units. one and another one Definition: So I Y'13S very very very busy in the episodic field .. - ftt t· t ~ f The \\vord out though really is for a new [slantlsm]!2071 'If/omen Is has alit n 0 IE pas In erm:; 0 been extremely hard. and irs worse novil thInk 'i Because 1started there vIas a great deal of equal opportunrty of public relatIOns going on ,And people ,\\-vere, I think. more conscious of the fact that there ViaS such a terrible inequaltty .A.nct they felt in some vrays that \"we had to hire one of them.\" ~them~ being \"v/om€n ~ .And. so. there were opportumtiss opening up, not. in any great !type 0111217Jv·rave of acceptance, but certainly. you know. small trickles of ~Iet·s g[\"e them a chance\" dialogue However thal\"s all past. and nobody feels that way about women anymore They say '\\vell, you know, we did hire them!\" so \"our conscious has been .I assuaged and we don't need to. you know, do anything about that anymore So [ think I1'S probably a lot harder, and J think that the independent ~iiiii~~iiiiiBiri01li·./SieiiPiroiPieirtiiesiii f\\i'/lmllh c'wirocmusenhadsiremcatonrisfe.satendd tIhthaitndkifwficoumtteynbydircercetaotrinsgnmeeadneyd samnaolluftillemt,syou know. an outlet, since they \"Jverenl getting the kind ohvork that they r.::-:------------Online Document: should have gotten, or should get And it's 1hat 't-tay to thiS day Doy Of course there are more women directing today. than they were when I = = - - . - - - - - - - - -Header: started. but there are more shm,vs. and more network. and more cable, and more everything No Header So the inequalities stili exist in term of the ratio of work versus ho\\-v many women \"''lark lind I donl r E3 Coding a Passage in the Interview in Joy's interview but now also in Berta'S and This procedure is only one of the many Queena's. With several nodes often attached to capabilities of this programo You may want to a single quote, the qualitative analysis allowed spend some time learning about this software me to find out which nodes were more likely before committing to ito What seems to be an to overlap with one another. efficient tool for me may not be for you. There is plenty qualitative research analysis software in One of my first observations was that the the market; try to find out what works for you. term sabotage was used fairly often by Joy and Queena I decided to run a report that Although it's important and appropriate to distin- would synthesize all the quotes that I attached guish between qualitative and quantitative re- to the node \"sabotage:' Figure 13-14 shows search, often to the point of discussing them sepa- the first page of the report created by NUD*IST rately, I don't want you to get the idea that they're The program searched for all quotes under \"sabotage,\" which is a subnode of \"discrimina- tion,\" for all online documents. It also provided the number assigned to each text unit, which allowed me to go back and see a quotation in the context of the whole document.

FIGURE 13-12 Creating an Index Tree Node:(1) people make it clear tMat you fdiSCriminatiOn ',vere a woman. and you ieU treated differently? J Yeah oM Y&-3h I mean... a lot of women hay-e commiserate about. you know ,\"\"hen you have to walk on Ihe sel for the firs I lime. they all used to workmg like a \"'fef! oiled machme and they say ·oh. here IS the vior:nan. somethmg dlr.erenl- and sometimes they can be horrible they can feslst),'OUf directing and the]' can, they can sabotage you, by laking a long limE 10 lIghl, or to mo.e sets, or to do something and doing that lime you're wasting Ilme, and that on a report, and the report goes to the [fronla!ll300] office, and, you and so on and so on and so on and so forth .And people upstairs donl knO\"#what the circumstances are. and they are .! fiGURE 13-13 Adding Two More Cases to the Analysis

400 Chapter 13: Qualitative Data Analysis _ I'J x .S.R,. HUD.IST Po..:er ver.n.on, re·Jl.Sl.on .... 0. x icensee: Sandrine :erb:1.1:I. Delete Coding POJEC1': Earl,_ User Sandrine Zerbib, 7:32 a:.:t., Dec 28~ 1999. Edit text unit <l I} / dl.s or iUl.nat loon/ swot .9.g'<1! ...... Ho Defi:u.tl.on !.argl.n c:odJ.!'lg l:eys for selected nodes: (1 1) /disc::l.:::::.inaticnfsahotage : +t ++++ ++ +++++ ++tH +t++t ++++ ttt +++++++++ Insert text unit Remove ++ OlI-LnrE DOCUI!EI11': Joy Spread . +t Doet.t:l.ent Header: eXamine coding 110 Header +t P.etn.. e>::al for th::.$' docu:::ent: l u...l.ts out: of 113, ;:; L8'\" .f !e::t u.....l.tS 100-101: \"iea..~. 011. yeah, I ::..ea...\"l ..... a lot of t!o::.en have cO=:!.$erate about, you A - .noT,)' vhen you have to ualk on the S€t tor the first t::.::.e, they all used A. A. o O'crk:::.n; ll.ke a usll ol.led l:lach:.ne and they say, '011.,_ here is the A. ·C:r.:ul..'1., so:.ething different ~ a...\"ld SOI:.1.el:l.::o.es the:\" ca.'\"} be horr~le. they c:a..\"l A., eSl.st your directing a.\"'l.d they c.an~ they can sabotage yeu, by taJang a le A ong ti:=.e to 1l.ght, or to =.ove sets, or to do sO!:l.'i!th::....\"l.g. a.'1.d dO:l.ng that 100 ). l.::::.e you' re uasting tl.ne, a.\"l.d that goes on a report, and the report goes A le o the [frontal] (3681 Offl.Ce,. a.\"\"\"l.d, you kno\"t:, &''1..1 so on :end so on and so .l- 101 A. n a.\"'\"l.d so forth. (1) Idl.scri:=.inat:l.on (1 1) Idiscn.::::'l.nat:ion/sabotage ;nd people upstairs don't: 1:no\"t: uhat the circu.:.:;stances are, and they are POt abOUt to fire a cine:::.atographer that ::'5 on the::.r shoy for e .... er a.'ld _....er ... nor do they \"lJa.\"\"\"l.t to knot:! that thl.s guy l.S a real bastard, a..\"'ld akl.ng your 1l. fe .9. horror. (1) ld.J..scri.=:.in.atl.on (1 1) /dl.scrl.:2.l.natl.on/sa.botage +t+t++++t++++++t++++++t+t++++++++++++++ : ++ OlI-LIm: DOCtiUE!I1: QUeena ++ DoctU:l.ent Header: No Header ++ P.eerie'Jal for tl'..is doC\\:::::.e;nt: 1S u..\"1-its out of 217, = 6.9\\ + 1e::t t.1.\"'\"l.l.tS 70-71: ut '!here uas a lot of H!SlStanc:e en the pa=t of t.he c:r€!U and there vas a A 70 A ot: of sabotage /discri:unation (1 1) Idiscri::oination/sabotage FIGURE 13-14 Analyzing the Node \"sabotage\" 800 o incompatible or competing. Unless you can operate in both modes, you'll limit your potential as a social 700 20 <Z> researcher. 40 ~ 600 In Chapter 14, I'll indicate some ways in which ~ 500 60 &\"- quantitative analyses can strengthen qualitative 80 :f!!::\" studies. Conversely, I want to end this chapter with § 400 an example of how quantitative data demand qual- 100 0 itative assessment. Z Figure 13-15 presents FBI data on the hour and 300 day of crimes committed in the United States 200 (Maltz 1998: 401). These data are often presented 100 in a tabular form, but notice how clearly the pat- terns of crime appear in this three-dimensional FIGURE 13-15 graph. The picture itself conveys the meaning of the statistical data. Summarizing it in the form of Number of One-an-One Homicides by Age of Vidim and Age equations-while possibly useful for certain pur- of Offender, Raw Data poses-adds nothing to the clarity of the picture. Source: Michael D.. Maltz,\"Visualizing Homocide: AResearch Note,\"}ournal of Indeed, there hardly seems a need to describe the pattern verbally. Here's a case where a picture is Quantitative Criminology 15, noA (1998):401. truly worth a thousand words.

Review Questions and Exercises 401 MAIN POINTS The Qualitative Analysis of Quantitative Data Introduction iii Qualitative analysis is the nonnumerical exami- iii Although qualitative and quantitative methods of analysis may appear incompatible or in com- nation and interpretation of observations. petition, research often demands that both kinds be used in the same project. Linking Theory and Analysis iii Qualitative analysis involves a continual inter- KEY TERMS play between theory and analysis. In analyzing The following terms are defined in context in the qualitative data, we seek to discover patterns chapter and at the bottom of the page where the term such as changes over time or possible causal is introduced, as well as in the comprehensive glossary links among variables. at the back of the book. iii Examples of approaches to the discovery and explanation of such patterns are Grounded axial coding memoing Theory Method (GTM), semiotics, and conver- case-oriented analysis open coding sation analysis. concept mapping qualitative analysis constant comparative selective coding Qualitative Data Processing method semiotics iii The processing of qualitative data is as much art conversation analysis (CA) variable-oriented cross-case analysis analysis as science. Three key tools for preparing data Grounded Theory for analysis are coding, memoing, and concept Method (GTM) mapping. iii In contrast to the standardized units used in REVIEW QUESTIONS AND EXERCISES coding for statistical analyses, the units to be coded in qualitative analyses may vary within 1. Review Goffman's examination of gender adver- a document. Although codes may be derived tising, then collect and analyze a set of advertis- from the theory being explored, more often re- ing photos, from magazines or newspapers, that searchers use open coding, in which codes are allow you to explore the relationship between suggested by the researchers' examination and gender and status. questioning of the data . iii Memoing is appropriate at several stages of 2. Review the discussion of homosexuality in the data processing to capture code meanings, Book of Leviticus and suggest ways that the ex- theoretical ideas, preliminary conclusions, amination might be structured as a cross-case and other thoughts that will be useful during analysis. analysis. iii Concept mapping uses diagrams to explore re- 3. Imagine you were conducting a cross-case lationships in the data graphically. analysis of revolutionary documents such as the Declaration of Independence and the Declara- Computer Programs for Qualitative Data tion of the Rights of Man and of the Citizen iii Several computer programs, such as NUD*IST, (from the French Revolution). Identify the key concepts you might code in the following are specifically designed to assist researchers sentence: in the analysis of qualitative data. In addition, researchers can take advantage of the capa- When in the Course of human events, it be- bilities of common software tools such as comes necessary for one people to dissolve word processors, database programs, and the political bands which have connected spreadsheets. them with another, and to assume among the Powers of the earth, the separate and equal station to which the Laws of Nature and of

402 Chapter 13: Qualitative Data Analysis Nature's God entitle them, a decent respect to McCormack, Coralie. 2004. \"Storying Stories: A the opinions of mankind requires that they Narrative Approach to In-Depth Interview should declare the causes which impel them Conversations.\" Intemational Journal of Social Re- to the separation. search Merhodology 7 (3): 219-36. The in-depth interviews common to qualitative field research 4. Write one code note and one theoretical note can result in lengthy narrative accounts that for Exercise 3. can pose daunting challenges for analysts. This article details a set of procedures for organizing 5. Using the library, InfoTrac College Edition, or the analysis of such stories, with a special the web, find a research report using conversa- concern for the ethical dimension. tion analysis. Summarize the main conclusions in your own words. Strauss, Anselm, and Juliet Corbin, eds. 1997. Grounded Theory in Practice. Thousand Oaks, ADDITIONAL READINGS CA: Sage. This updated statement of grounded theory offers special guidance on coding and Berg, Bruce. 1998. Qualitarive Research Methods for memoing. the Social Sciences, Boston: Allyn and Bacon. Here's a comprehensive and readable review Strauss, Anselm, and Juliet Corbin. 1998. Basics of of the techniques for collecting and analyzing Qualitative Research: Techniques and Procedures for qualitative data, with a special sensitivity to Developing Grounded TheOlY. Thousand Oaks, research ethics, CA: Sage. Denzin, Norman re, and Yvonna S, Lincoln. 1994. SPSS EXERCISES Handbook of Qualirative Research. Thousand Oaks, See the booklet that accompanies your text for CA: Sage, Here's a rich resource covering many exercises using SPSS (Statistical Package for the aspects of qualitative research, in both theory Social Sciences). There are exercises offered for each and practice. chapter, and you'll also find a detailed primer on using SPSS. Glaser, Barney G., and Anselm L Strauss. 1967. The Discovery of Grounded Theory.: Strategies for Qualita- Online Study Resources rive Research, Chicago: Aldine. This is the classic statement of grounded theory, with practical Sociology~Now'M: Research Methods suggestions that are still useful today. I.. Before you do your final review of the chapter, Hutchby, Ian, and Robin Wooffitt. 1998. COl1versarioll take the SociologyNow: Research iVlethods diagnos- Analysis: Principles, Practices and Applications. tic quiz to help identify the areas on which you Cambridge, England: Polity Press. An excellent should concentrate. You'll find information on overview of the conversation analysis method. this online tool, as well as instructions on how The book examines the theory behind the to access all of its great resources, in the front of technique, how to use it. and some possible the book. applications. 2.. As you review, take advantage of the Sociology Jacobson, David. 1999. \"Doing Research in Cyber- Now: Research Methods customized study plan, space. \" Field Methods II (2): 127-45. The use based on your quiz results. Use this study plan of the Internet for social research is not limited with its interactive exercises and other re- to surveys and experiments, as Jacobson dem- sources to master the materiaL onstrates in this examination of computer- mediated communication (CMC). 3. When you're finished with your review, take the posttest to confirm that you're ready to King, Gary, Robert 0 . Keohane, and Sidney Verba. move on to the next chapter. 1994. Designing Social Inquiry: Scient({ic Inference in Qualitative Research. Princeton, NJ: Princeton University Press. This controversial book by three political scientists seeks to bring the logic of causal, quantitative analysis to bear on quali- tative data. Their stated intention is to unify the two approaches.

Online Study Resources 403 WEBSITE FOR THE PRACTICE Some Computer Programs for Qualitative OF SOCIAL RESEARCH 11TH EDITION Analysis http://www.ideaworksocom/Qualrus.shtml Go to your book's website at http://sociology Qualrus .wadsworth.com/babbie_practicelle for tools to aid you in studying for your exams. You'll find Tmo- http://wwvv.qsLcom.au/ rial Quizzes with feedback, Internet Exercises, Flashcards, l\\'UD*IST, N4, NVivo, N5 and Chapter Tworials, as ,veil as E)aended Projects. IlIfo- hac College Edition search terms, Social Research in Cyber- http://wwwqualisresearchocom/info.htm space, GSS Data, Web Links, and primers for using vari- Ethnograph ous data-analysis software such as SPSS and NVivo. http://wwwoatlastLde/ WEB LINKS FOR THIS CHAPTER Atlas.ti Please realize that the Internet is an evolv- http://www.researmware. com/ ing entity, subject to change. Nevertheless, HyperResearch these few websites should be fairly stable. Also, check your book's website for even more Web Judy Norris, Extensive Listing of Qualitative Links These websites, current at the time of t11is book's Data Analysis Programs publication, provide opportunities to learn about qual- http://w,'vw.qualitativeresearch.uga.edu IQualPagel itative research and analysis. This website provides hot links to additional qualita- tive data analysis programs and illustrates the range of FOIum: Qualitative Social Research possibilities for computer analyses of qualitative data. http://vvww.qualitative-research.net/fqs/fqs-eng.htm This is a peer-reviewed, multilinguistic journal dealing with all aspects of qualitative research, with abstracts and articles you can read online.

Quantitative Data Analysis Introduction Subgroup Comparisons \"Collapsing\" Response Quantification of Data Categories Developing Code Handling \"Don't Knows\" Categories Numerical Descriptions Codebook Construction in Qualitative Research Data Entry Bivariate Analysis Univariate Analysis Percentaging a Table Distributions Constructing and Reading Central Tendency Bivariate Tables Dispersion Continuous and Discrete Introduction to Multivariate Variables Analysis Detail versus Manageability Sociological Diagnostics Sociology@Now'M: Research Methods Use this online tool to help you make the grade on your next exam. After reading this chapter, go to the \"Online Study Resources\" at the end of the chapter for instructions on how to benefit from SoaologyNow: Research Methods.

Quantification of Data 405 Introduction rocket science. Researchers can also easily assign numerical representations to such variables as reli- In Chapter 13, we saw some of the logic and tech- giolls affiliation, political party, and region ofthe country. niques by which social researchers analyze the qualitative data they've collected. This chapter ex- Some data are more challenging, however. If a arnines quantitative analysis, or the techniques survey respondent tells you that he or she thinks by which researchers convert data to a numerical the biggest problem facing Woodbury, Vermont to- form and subject it to statistical analyses. day is \"the disintegrating ozone layer,\" the com- puter can't process that response numerically. You To begin we'll look at quantification-the pro- must translate by coding the responses. We've al- cess of converting data to a numerical format. This ready discussed coding in connection with content involves converting social science data into a analysis (Chapter 11) and again in connection lvith machine-readable fonn-a form that can be read and qualitative data analysis (Chapter 13). Now we look manipulated by computers and similar machines at coding specifically for quantitative analysis. used in quantitative analysis. To conduct a quantitative analysis, researchers The rest of the chapter ,viII present the logic often must engage in a coding process after the and some of the techniques of quantitative data data have been collected. For example, open-ended analysis-starting lvith the simplest case, univari- questionnaire items result in nonnumerical re- ate analysis, which involves one variable, then sponses, which need to be coded before analysis. discussing bivariate analysis, which involves two As with content analysis, the task is to reduce a variables. We'll end lvith a brief introduction to ,vide variety of idiosyncratic items of information to multivariate analysis, or the examination of several a more limited set of attributes composing a vari- variables simultaneously, such as age, education, and able. Suppose, for example, that a survey re- prejudice. searcher asks respondents, \"What is your occupa- tion?\" The responses to such a question \\viII vary Before we can do any sort of analysis, we need considerably. Although he or she can assign a sepa- to quantify our data. Let's turn now to the basic rate numerical code to each reported occupation, steps involved in converting data into machine- this procedure lvill not facilitate analysis, which readable forms amenable to computer processing typically depends on several subjects having the a:nd analysis. same attribute. Quantification of Data The variable occupation has many preestablished coding schemes. One such scheme distinguishes Today, quantitative analysis is almost always professional and managerial occupations, clerical handled by computer programs such as SPSS and occupations, semiskilled occupations, and so forth. MicroCase. For those programs to work their Another scheme distinguishes different sectors of magic, they must be able to read the data you've the economy: manufacturing, health, education, collected in your research. If you've conducted a commerce, and so forth. Still others combine both. survey, for example, some of your data are inher- Using an established coding scl1eme gives you the ently numerical: age or income, for example. advantage of being able to compare your research Whereas the writing and check marks on a ques- tionnaire are qualitative in nature, a scribbled age quantitative analysis The numerical representa- is easily converted to quantitative data. tion and manipulation of observations for the pur- pose of describing and eli:plaining the phenomena Other data are also easily quantified: trans- that those observations reflect fonning male and female into \"I\" and \"2\" is hardly

406 Chapter 14: Quantitative Data Analysis results ,\"lith those of other studies. (See for instance TABLE 14-1 the Bureau of Labor Statistics at http://stats.bls.gov/ Student Responses That Can Be Coded \"Financial Concerns\" soc/socmajo.htm. ) Tuition is too high Financial Concerns The occupational coding scheme you choose Not enough parking spaces should be appropriate to the theoretical concepts Faculty don't know what they are doing x being examined in your study. For some studies, Advisors are never available coding all occupations as either white-collar or Not enough classes offered x blue-collar might be sufficient. For others, self- Cockroaches in the dorms employed and not self-employed might do. Or a Too many requirements X peace researcher might '''lish to know only whether Cafeteria food is infected the occupation depended on the defense establish- Books cost too much ment or not. Not enough financial aid Although the coding scheme should be tailored Advisors are never available to meet particular requirements of the analysis, one general guideline should be kept in mind. If the Not enough classes offered data are coded to maintain a great deal of detail, code categories can always be combined during an Cockroaches in the dorms analysis that does not require such detaiL If the data are coded into relatively fevv, gross categories, Too many requirements however. there's no way during analysis to recreate the original detail. To keep your options open, it's a Cafeteria food is infected good idea to code your data in greater detail than you plan to use in the analysis. Books cost too much Developing Code Categories Not enough financial aid There are two basic approaches to the coding pro- Take a minute to review these responses and see cess. First, you can begin with a relatively well- whether you can identify some categories repre- developed coding scheme. You may choose to do sented. Realize that there is no right answer; there this because it serves your research purpose. Thus, are several coding schemes that might be generated as suggested previously, the peace researcher might from these answers. code occupations in terms of their relationship to the defense establishment. Or. you may want to Let's start with the first response: \"Tuition is too use an existing coding scheme because it allows high.\" What general areas of concern does that re- you to compare your findings with those of previ- sponse reflect? One obvious possibility is \"Financial ous research. Concerns.\" Are there other responses that would fit into that category? Table 14-1 shows which of the Second, you can generate codes from your questionnaire responses could fit into that category. data, as discussed in Chapter 13. Let's say we've asked students in a self-administered campus sur- In more general terms, the first answer can vey to state what they believe is the biggest prob- also be seen as reflecting nonacademic concerns. lem facing their college today. Here are a few of the This categorization would be relevant if your re- answers they might have written in. search interest included the distinction between academic and nonacademic concerns. If that were Tuition is too high the case, the responses might be coded as shown in Not enough parking spaces Table 14-2. Faculty don't know what they are doing Notice that I didn't code the response \"Books cost too much\" in Table 14-2, because this concern could be seen as representing both of the categories.

Quantification of Data 407 TABLE 14-2 TABLE 14-3 Student Concerns Coded as \"Academic\" and \"Nonacademic\" Nonacademic Concerns Coded as \"Administrative\" or \"Facilities\" Academic Nonacademic Tuition is too high X Academic Administrative Facilities Not enough parking spaces X Faculty don't know what Tuition is too high X they are doing X Advisors are never available X Not enough parking spaces X Not enough classes offered X Cockroaches in the dorms Faculty don't know what X Too many requirements X they are doing Cafeteria food is infected X Books cost too much Advisors are never available X Not enough financial aid X Not enough classes offered X X Cockroaches in the dorms X Too many requirements X Cafeteria food is infected X Books cost too much X Not enough financial aid X Books are part of the academic program, but their exhaustive and mutually exclusive. Every piece of cost is not. This signals the need to refine the cod- information being coded should fit into one and ing scheme we're developing. Depending on our re- only one category. Problems arise whenever a search purpose, we might be especially interested in given response appears to fit equally into more identifying any problems that had an academic ele- than one code category or whenever it fits into no ment; hence we'd code this one \"Academic.\" Just as category: Both signal a mismatch between your reasonably, however. we might be more interested data and your coding scheme. in identifying nonacademic problems and would code the response accordingly. Or, as another alter- If you're fortunate enough to have assistance in native, we might create a separate category for re- the coding process, you'll need to train your coders sponses that involved both academic and nonacad- in the definitions of code categories and show them emic matters. how to use those categories properly. To do so, ex- plain the meaning of the code categories and give As yet another alternative, we might want to several examples of each. To make sure your coders separate nonacademic concerns into those involv- fully understand what you have in mind, code sev- ing administrative matters and those dealing '''lith eral cases ahead of time. Then ask your coders to campus facilities. Table 14-3 shows how the first code the same cases vvithout knowing how you ten responses would be coded in that event. coded them. Finally, compare your coders' work with your own. Any discrepancies will indicate an As these few examples illustrate, there are imperfect communication of your coding scheme to many possible schemes for coding a set of data. your coders. Even ''lith perfect agreement between Your choices should match your research purposes you and your coders, however, it's best to check the and reflect the logic that emerges from the data coding of at least a portion of the cases throughout themselves. Often, you'll find yourself modifying the coding process. the code categories as the coding process proceeds. Whenever you change the list of categories, how- If you're not fortunate enough to have assis- ever, you must review the data already coded to tance in coding, you should still obtain some see whether changes are in order. verification of your own reliability as a coder. No- body's perfect, especially a researcher hot on the Like the set of attributes composing a variable, trail of a finding. Suppose that you're studying an and like the response categories in a closed-ended questionnaire item, code categories should be both

408 III Chapter 14: Quantitative Data Analysis POLVIEWS ATTEND We hear a lot of talk these days about liberals and conserva- How often do you attend religious services? tives.I'm going to show you aseven-point scale on which the political views that people might hold are arranged from ex- o. Never tremely liberal-point l-to extremely conservative- point 7.Where would you place yourself on this scale? 1. Less than once ayear 2. About once or twice ayear 1. Extremely liberal 3. Several times ayear 4. About once a month 2. liberal 5. 2-3 times a month 3. Slightly liberal 6. Nearly every week 7. Everyweek 4. Moderate, middle ofthe road 8. Several times aweek 5. Slightly conservative 9. Don't know, No answer 6. Conservative 7. Extremely conservative 8. Don't know 9. No answer FIGURE 14-1 APartial Codebook emerging cult and that you have the impression file. A codebook is a document that describes the that people who do not have a regular family will locations of variables and lists the assignments of be the most likely to regard the new cult as a fam- codes to the attributes composing those variables. ily substitute. The danger is that whenever you dis- cover a subject who reports no family, you'll un- A codebook serves two essential functions. consciously try to find some evidence in the First, it's the primary guide used in the coding pro- subject's comments that the cult is a substitute for cess. Second, it's your guide for locating variables family. If at all possible, then, get someone else to and interpreting codes in your data file during code some of your cases to see whether that person analysis. If you decide to correlate two variables as makes the same assignments you made. a part of your analysis of your data, the codebook tells you where to find the variables and what the Codebook Construction codes represent. The end product of the coding process is the con- Figure 14-1 is a partial codebook created from version of data items into numerical codes. These two variables from the General Social Survey. Al- codes represent attributes composing variables, though there is no one right format for a codebook, which, in turn, are assigned locations within a data this example presents some of the common elements. codebook The document used in data processing and analysis that tells the location of different data Notice first that each variable is identified by an items in a data file. Typicall}, the codebook identi- abbreviated variable name: POLVIEWS, ATTEND. fies the locations of data items and the meaning of We can determine the religious services attendance the codes used to represent different attributes of of respondents, for example, by referencing AT- variables. TEND. This example uses the format established by the General Social Survey, which has been carried over into SPSS. Other data sets and/or analysis pro- grams might format variables differently. Some use numerical codes in place of abbreviated names, for

Univariate Analysis \" 409 example. You must however, have some identifier If your data have been collected by question- that will allow you to locate and use the variable in naire, you might do your coding on the question- question. naire itself. Then, data entry specialists (including yourself) could enter the data into, say, an SPSS Next, every codebook should contain the full data matrix or into an Excel spreadsheet to be im- definition of the variable. In the case of a question- ported later into SPSS. naire, the definition consists of the exact wordings of the questions asked, because, as we've seen, the Sometimes social researchers use optical scan wording of questions strongly influences the an- sheets for data collection. These sheets can be fed swers returned. In the case of POLVlEWS, you into machines that convert the black marks into know that respondents were handed a card con- data, which can be imported into the analysis pro- taining the several political categories and asked to gram. This procedure only works with subjects pick the one that best fit them. who are comfortable using such sheets, and it's usually limited to closed-ended questions. The codebook also indicates the attributes com- posing each variable. In POLVlEWS, for example, Sometimes, data entry occurs in the process of respondents could characterize their political orien- data collection. In Computer Assisted Telephone tations as \"Extremely libera\\,\" \"Libera\\,\" \"Slightly Interviewing, for example, the interviewer keys re- liberal,\" and so forth. sponses directly into the computer, where the data are compiled for analysis (see Chapter 9). Even Finally, notice that each attribute also has a nu- more effortless, online surveys can be constructed meric labeL Thus, in POLVlEWS, \"Extremely lib- so that the respondents enter their own answers di- eral\" is code category 1. These numeric codes are rectly into the accumulating database, without the used in various manipulations of the data. For ex- need for an intervening interviewer or data entry ample, you might decide to combine categories 1 person. through 3 (all the \"liberal\" responses). It's easier to do this with code numbers than with lengthy Once data have been fully quantified and en- names. tered into the computer, researchers can begin quantitative analysis. Let's look at the three cases You can visit the GSS codebook online at mentioned at the start of this chapter: univariate, http://webappjcpsLumich.edu/GSS/. If you know bivariate, and multivariate analyses. the symbolic name (e.g., POLVIEWS), you can lo- cate it in the Mnemonic listing. Otherwise, you can Univariate Analysis browse the \"Index by Subject\" to find all the differ- ent questions that have been asked regarding a par- The simplest form of quantitative analysis, univari- ticular topic. ate analysis, involves describing a case in terms of a single variable-specifically, the distribution Data Entry of attributes that it comprises. For example, if gen- der were measured, we would look at how many In addition to transforming data into quantitative of the subjects were men and how many were form, researchers interested in quantitative analysis women. also need to convert data into a machine-readable format, so that computers can read and manipulate univariate analysis The analysis of a single vari- the data. There are many ways of accomplishing able, for purposes of description. Frequency distribu- this step, depending on the original form of your tions, averages, and measures of dispersion would be data and also the computer program you choose for examples of univariate analysis, as distinguished analyzing the data. I'll simply introduce you to the from bivariate and multivariate analysis. process here. If you find yourself undertaking this task, you should be able to tailor your work to the particular data source and program you're using.

410 Chapter 14: Quantitative Data Analysis TABLE 14-4 Services, 2000 GSS Attendance at Act-end How Often R Attends Religious Services \\!alUe Label Valid Cum LT ONCE X·\\. YEP-_R ONCE A YEAR Value Percent Percent Percent SEVRL TIHES A YR 0 ONCE -'c,. IvIONTH 1 583 20.7 21.3 21.3 2 2-3X A MONTH 3 219 7.8 8.0 29.3 4 E\\IERY T,lJEEIZ 5 334 11.9 12.2 41.5 MORE THN ONCE VJK 6 DK,NA 7 369 13 .1 13.5 55.0 8 9 197 7.0 7.2 62.2 Total 221 7.8 8.1 70.3 131 4.7 4.8 75.0 488 17.3 17.8 92.9 195 6.9 7.1 100.0 80 2.8 Missing 2,817 100.0 100.0 Valid cases 2,737 Hissing cases 80 Source: General Social Survey, 2000 Distributions By analogy, suppose your best friend tells you that they drank a six-pack of beer. Is that a little The most basic format for presenting univariate data beer or a lot? The answer, of course, depends on is to report all individual cases, that is, to list the at- whether they consumed the beer in a month, a tribute for each case under study in terms of the week, a day, or an hour. In the case of religious variable in question. Let's take as an example the participation, similarly, we need some basis for as- General Social Survey (GSS) data on attendance at sessing the number that represents the people who religious services, ATTEND. Table 14A presents the never attend religious services. results of an SPSS analysis of this variable. One way to assess the number is to calculate Let's examine the table, piece by piece. First if the percentage of all respondents who said they you look near the bottom of the table, you'll see never go to religious services. If you were to divide that the sample being analyzed has a total of 2,817 583 by the 2,737 who gave some answer, you cases. In the last row above the totals, you'll see would get 21.3 percent, which appears in the table that 80 of the 2,817 respondents either said they as the Valid Percent. Now we can say that 21 per- didn't know (DK) or gave no answer (NA) in re- cent, or roughly one U.S. adult in five, reports sponse to this question. So our assessment of U.S. never attending religious services. attendance at religious services in 2000 will be based on the 2,737 respondents who answered the This result is more meaningfuL but does it sug- question . gest that people in the United States are generally nonreligious? A further look at Table 14-4 shows Go back to the top of the table now. You'll see that the response category most often chosen was that 583 people said they never went to religious Every Week, with 17.8 percent of the respondents services. This number in and of itself tells us noth- giving that answer. Add to that the 7.1 percent ing about religious practices. It does not, in itself, who report attending religious services more than give us an idea of whether the average American once a week, and we find that nearly a fourth attends religious services a little or a lot. (24.9 percent) of U.S. adults say they attend

Univariate Analysis 411 10 1------£ 0r J Never Less than Once Several Once 2-3 J~. Every More Missing once a year times a month week than once a year a year times Nearly a month a week every week How often R attends religious services FIGURE 14-2 Bar Chart of GSS ATIEND, 2000 religious services at least once a week. As you can go through extended periods without getting any see, each new comparison gives a more complete hits at all and go through other periods when he picture of the data. or she gets a bunch of hits all at once. Over time, though, the central tendency of the batter's perfor- A description of the number of times that the mance can be expressed as getting three hits in various attributes of a variable are observed in a every ten chances. Similarly, your grade point aver- sample is called a frequency distribution. Some- age expresses the \"typical\" value of all your grades times it's easiest to see a frequency distribution in a taken together, even though some of them might graph. Figure 14-2 was created by SPSS from the be A's, others B's, and one or two might be C's (I GSS data on ATTEND. The vertical scale on the left know you never get anything lower than a C). side of the graph indicates the percent selecting each of the answers displayed along the horizontal Averages like these are more properly called axis of the graph. Take a minute to notice how the the arithmetic mean (the result of dividing the percentages in Table 14-4 correspond to the heights of the bars in Figure 14-2. frequency distribution A description of the num- ber of times the various attributes of a variable are ob- Central Tendency served in a sample. The report that 53 percent of a sample were men and 47 percent were women would Beyond sinlply reporting the overall distribution of be a simple example of a frequency distribution. values, sometinles called the marginal frequencies or just the marginals, you may choose to present your average An ambiguous teml generally suggesting data in the form of an average or measure of een- typical or normal-a central tendency. The mean, Hal tendency. You're already familiar with the con- median, and mode are specific examples of mathe- cept of central tendency from the many kinds of matical averages. averages you use in everyday life to express the \"typical\" value of a variable. For instance, in base- mean An average computed by summing the val- ball a batting average of .300 says that a batter gets ues of several observations and dividing by the num- a hit three out of every ten opportunities-on ber of observations. If you now have a grade point average. Over the course of a season, a hitter might average of 4.0 based on 10 courses, and you get an F in this course, your new grade point (mean) average vl'ill be 3.6.

412 III Chapter 14: Quantitative Data Analysis sum of the values by the total number of cases)., each age by the number of subjects who have that The mean is only one way to measure central ten- age, (2) total the results of all those multiplications, dency or \"typical\" values. Two other options are and (3) divide that total by the number of subjects. the mode (the most frequently occurring attri- bute) and the median (the middle attribute in the In the case of age. a special adjustment is ranked distribution of observed attributes). Here's needed. As indicated in the discussion of the mode, how the three averages would be calculated from a those who call themselves \"13\" actually range from set of data. exactly 13 years old to those just short of 14. It's reasonable to assume. moreover. that as a group Suppose you're conducting an ex,})eriment that the\" 13-year-olds\" in the country are evenly dis- involves teenagers as subjects. They range in age tributed within that one-year span, making their from 13 to 19, as indicated in the following table: average age 13.5 years. This is true for each of the age groups. Hence, it is appropriate to add 0.5 years Age Number to the final calculation, making the mean age 16.37. as indicated in Figure 14~3., 13 3 14 4 The third measure of central tendency. the me- 15 6 dian. represents the \"middle\" value: Half are above 16 8 it. half below. If we had the precise ages of each 17 4 subject (for example. 17 years and 124 days). we'd 18 3 be able to arrange all 31 subjects in order by age, 19 3 and the median for the whole group would be the age of the middle subject. Now that you've seen the actual ages of the 31 subjects, how old would you say they are in gen- As you can see. however. we do not know pre- eraL or \"on average\"? Let's look at three different cise ages; our data constitute \"grouped data\" in this ways you might answer that question. regard. For example. three people who are not pre- cisely the same age have been grouped in the cate- The easiest average to calculate is the mode, the gory \"13-year~olds.\" most frequent value. As you can see, there were more 16-year-olds (eight of them) than any other Figure 14-3 illustrates the logic of calculating a age, so the modal age is 16, as indicated in Figure median for grouped data. Because there are 31 sub- 14-3. Technically, the modal age is the category jects altogether. the \"middle\" subject would be sub- \"16,\" which may include some people who are ject number 16 if they were arranged by age-IS closer to 17 than 16 but who haven't yet reached teenagers would be younger and 15 older. Look at that birthday. the bottom portion of Figure 14-3, and you'll see that the middle person is one of the eight 16-year- Figure 14-3 also demonstrates the calculation olds. In the enlarged view of that group, we see of the mean. There are three steps: (1) multiply that number 16 is the third from the left. mode An average representing the most frequently Because we do not know the precise ages of observed value or attribute. If a sample contains the subjects in this group. the statistical convention LOOO Protestants, 275 Catholics. and 33 Jews, here is to assume they are evenly spread along the Protestant is the modal category. width of the group. In this instance, the possible median An average representing the value of the ages of the subjects go from 16 years and no days \"middle\" case in a rank-ordered set of observations. to 16 years and 364 days. Strictly speaking. the If the ages of five men are 16, 17. 20, 54. and 88, the range, then, is 364/365 days. As a practical matter, median would be 20. (The mean would be 39.) it's sufficient to call it one year. If the eight subjects in this group were evenly spread from one limit to the other, they would be one-eighth of a year apart from each other-a 0. 125-year intervaL Look at the illustration and

Age Number Mode = 16 Most frequent itt titt tttttt t t t t t i <Qw+-..~.., \"\", tttt itt ttt Age Number ttt 13 x 3 =39 tttt 14 x 4 = 56 ittttt 15 x 6 =90 titttitt 16 x 8 =128 tttt 17 x 4 = 68 Mean = 16,37 tit 18x3=54 Arithmetic average ttt 19 x 3 = 57 ~ 492 ~ 31 = 15.,87 + 0.50 =16.37 Age Number (Total) (Cases) r Median = 16.31 Midpoint ttt 1-3 tttt 4-7 tttttt 8-13 14 15 16 17 18 19 20 21 tttttttt tttt 22-25 t t tt t t tf•'~'$> 16.06 16.19 16,31 16.44 16.56 16. 69 16.81 16.94 ttt 26-28 29-31 ttt FIGURE 14-3 Three \"Averages\"

414 Chapter 14: Quantitative Data Analysis you'll see that if we place the first subject half the Washington, has a net worth in excess of a million interval from the lower limit and add a full interval dollars. ff you were to visit Redmond, however. you to the age of each successive subject, the final one would not find that the \"average\" resident lives up is half an interval from the upper limit. to your idea of a millionaire. The very high mean reflects the influence of one extreme case among What we've done is calculate, hypothetically, Redmond's 40,000 residents-Bill Gates of MicTO- the precise ages of the eight subjects-assuming soft, who has a net worth (at the time this is being their ages were spread out evenlyo Having done this, written) of tens of billions of dollars. Clearly. the we merely note the age of the middle subject- median wealth would give you a more accurate pic- 1631-and that is the median age for the group 00 ture of the residents of Redmond as a whole. Whenever the total number of subjects is an This example should illustrate the need to even number. of course, there is no middle case. choose carefully among the various measures of To get the median, you merely calculate the mean central tendency. A course or textbook in statistics of the two values on either side of the midpoint in will give you a fuller understanding of the variety the ranked data. Suppose, for example, that there of situations in which each is appropriate. was one more 19-year-old in our sample, giving us a total of 32 caseso The midpoint would then fall be- Dispersion tween subjects 16 and 17. The median would there- Averages offer readers the advantage of reducing fore be calculated as (1631 + 16.44)/2 = 16.38. the raw data to the most manageable form: A single number (or attribute) can represent all the detailed As you can see in Figure 14- 3, the three mea- data collected in regard to the variable. This advan- sures of central tendency produce three different tage comes at a cost, of course, because the reader values for our set of data, which is often (but not cannot reconstruct the original data from an aver- necessarily) the case. Which measure, then, best age. Summaries of the dispersion of responses can represents the \"typical\" value? More generally, somewhat alleviate this disadvantage. which measure of central tendency should we pre- fer? The answer depends on the nature of your data Dispersion refers to the way values are distrib- and the purpose of your analysis. For example, uted around some central value, such as an aver- whenever means are presented, you should be age. The simplest measure of dispersion is the aware that they are susceptible to extreme values- range: the distance separating the highest from the a few very large or very small numbers. As only one lowest value. Thus, besides reporting that our sub- example, the (mean) average person in Redmond, jects have a mean age of 15.87, we might also indi- cate that their ages range from 13 to 19. dispersion The distribution of values around some central value, such as an averageo The range is a A more sophisticated measure of dispersion is simple example of a measure of dispersion. Thus, we the standard deviation. This measure was briefly may report that the mean age of a group is 37.9, and mentioned in Chapter 7 as the standard error of a the range is from 12 to 89. sampling distribution. Essentially, the standard de- viation is an index of the amount of variability in a standard deviation A measure of dispersion set of data. A higher standard deviation means that around the mean, calculated so that approximately the data are more dispersed; a lower standard devi- 68 percent of the cases will lie within plus or minus ation means that they are more bunched together. one standard deviation from the mean, 95 percent Figure 14-4 illustrates the basic ideao Notice that the will lie within plus or minus two standard devia- professional golfer not only has a lower mean score tions, and 99.9 percent will lie within three standard but is also more consistent-represented by the deviations. Thus, for example, if the mean age in a smaller standard deviation. The duffer, on the group is 30 and the standard deviation is 10, then other hand, has a higher average and is also less 68 percent have ages between 20 and 40. The smaller the standard deviation, the more tightly the values are clustered around the mean; if the standard devia- tion is high, the values are widely spread OllL

a. High standard deviation = spread-out values Univariate Analysis 415 Amateur Golfer's Scores and discrete. A continuous variable (or ratio vari- able) increases steadily in tiny fractions\" An example (f) is age, which increases steadily with each increment Q) of time. A discrete variable jumps from category to category without intervening steps. Examples in- Eco clude gender, lIlilirmy rallk, and year ill college (you go from being a sophomore to a junior in one step). OJ In analyzing a discrete variable-a nominal or '0 ordinal variable, for example-some of the tech- ..OaJ niques discussed previously do not apply. Strictly E speaking, modes should be calculated for nominal data, medians for interval data, and means for ratio :J data, not for nominal data (see Chapter 5). If the variable in question is gellder, for example, raw Z numbers (23 of the cross-dressing outlaw bikers in our sample are women) or percentages (7 percent '------y-----' are women) can be appropriate and useful analy- 68% of values ses, but neither a median nor a mean would make any sense. Calculating the mode would be legiti- b. Low standard deviation = tightly clustered values mate, though not very revealing, because it would only tell us \"most were men.\" However. the mode Professional Golfer's Scores for data on religious affiliation might be more inter- esting, as in \"most people in the United States are (f) Protestant.\" Q) Detail versus Manageability Eco In presenting univariate and other data, you'll be OJ constrained by two goals. On the one hand, you should attempt to provide your reader 'with the '0 ..OaJ continuous variable A variable whose attributes E form a steady progression, such as age or income. Thus, the ages of a group of people might include :J 21. 22, 23, 24, and so forth and could even be bro- ken down into fractions of years. Contrast this with Z discrete variables, such as gender or religious affiliation, whose attributes form discontinuous chunks. 68% of values discrete variable A variable whose attributes are separate from one another, or discontinuous, as in FIGURE 14-4 the case of gender or religioliS affiliation. Contrast this High and Low Standard Deviations with continuous variables, in which one attribute shades off into the neXL Thus, in age (a continuous consistent: sometimes doing much better, some- variable), the attributes progress steadily from 21 to times much worse. 22 to 23, and so forth, whereas there is no progres- sion from male to female in the case of gender. There are many other measures of dispersion. In reporting intelligence test scores, for example, researchers might determine the interquartile ral1ge, the range of scores for the middle 50 percent of subjects. If the top one-fourth had scores ranging from 120 to 150, and if the bottom one-fourth had scores ranging from 60 to 90, the report might say that the interquartile range was from 90 to 120 (or 30 points) with a mean score of. let's say, 102. Continuous and Discrete Variables The preceding calculations are not appropriate for all variables. To understand this point. we must dis- tinguish between two types of variables: continuous

416 \" Chapter 14: Quantitative Data Analysis TABLE 14-5 TABLE 14-6 Marijuana Legalization by Age of Respondents, 2000 Marijuana Legalization by Political Orientation, 2000 55 and 5hould Should Under 27 21-35 36-54 aldel Legalize Nat Legalize 100%= Should be legalized 42% 42% 35% 23% Extremely liberal 55% 45 (69) Should not be legalized 58 58 65 77 Liberal 54% 46 (199) (64) (500) (715) (501) Slightly liberal 41% 59 (172) 100% = Moderate 32% 68 (649) Slightly conservative 30% 70 (244) Source: General Social Survey, 2000, Conservative 20% 80 (280) Extremely conservative 25% 75 (57) fullest degree of detail regarding those data. On the other hand, the data should be presented in a man- Source: General Social Survey, 2000. ageable form. As these two goals often directly counter each other, you'll find yourself continually First, let's see how another set of subgroups an- seeking the best compromise between them. One swered this question. useful solution is to report a given set of data in more than one form. In the case of age, for example, Table 14-6 presents different political subgroups' you might report the distribution of ungrouped ages attitudes toward legalizing marijuana, based on pillS the mean age and standard deviation. whether respondents characterized themselves as conservative or liberaL Before looking at the table, As you can see from this introductory discus- you might try your hand at hypothesizing what the sion of univariate analysis, this seemingly simple results are likely to be and why. Notice that I've matter can be rather complex. In any event, the les- changed the direction of percentaging this table, to sons of this section pave the way for a consideration make it easier to read. To compare the subgroups in of subgroup comparisons and bivariate analyses. this case, you would read down the columns, not Subgroup Comparisons across them. Before examining the logic of causal analysis, Univariate analyses describe the units of analysis of a study and, if they are a sample drawn from some let's consider another example of subgroup com- larger population, allow us to make descriptive in- parisons: one that will let us address some table- ferences about the larger population. Bivariate and formatting issues. multivariate analyses are aimed primarily at expla- nation. Before turning to explanation, however, we \"Collapsing'Response Categories should consider the case of subgroup description. \"Textbook examples\" of tables are often simpler Often it's appropriate to describe subsets of than you'll typically find in published research re- cases, subjects, or respondents. Here's a simple ex- ports or in your own analyses of data, so this sec- ample from the General Social Survey. In 2000, re- tion and the next one address two common prob- spondents were asked, \"Should marijuana be made lems and suggest solutions. legal?\" In response, 33.5 percent said it should and 66.5 percent said it shouldn't. Table 14-5 presents Let's begin by turning to Table 14-7, which re- the responses given to this question by respondents ports data collected in a multinational poll con- in different age categories\" ducted by the New York Times, CBS News, and the Herald Tribune in 1985, concerning attitudes about Notice that the subgroup comparisons tell us the United Nations. The question reported in Table how different groups in the population responded 14-7 deals with general attitudes about the way the to this question. You can undoubtedly see a pattern UN was handling its job. in the results; we'll return to that in a moment.

TABLE 14-7 Subgroup Comparisons\" 417 Attitudes toward the United Nations: \"How is the UN doing in solving the problems it has had to face?\" Japan United States 1% 5% West Gelmany Britain Frame 11 46 2% 43 27 Very good job 2% 7% 45 5 13 Good job 46 39 22 41 10 Poor job 21 28 3 Very poor job 69 28 Japan United States Don't know 26 17 12% 51% 48 40 Source: \"S-Nation Survey Finds Hope for U.N.,\" New York Times, June 26, 1985, p. 6. 41 10 TABLE 14-8 Collapsing Extreme Categories West Germany Britain France 46% 47% Good job or better 48% 37 25 Poor job or worse 27 17 28 Don't know 26 Here's the question: How do people in the five combine \"very good\" with \"good\" and \"very poor\" nations reported in Table 14-7 compare in their with \"POOL\" If you were to do this in the analysis support for the kind of job the UN was doing? As of your own data, it would be ,vise to add the raw you review the table, you may find there are sim- frequencies together and recompute percentages ply so many numbers that it's hard to see any for the combined categories, but in analyzing a meaningful pattern. published table such as this one, you can simply add the percentages as illustrated by the results Part of the problem with Table 14-7 lies in the shown in Table 14-8. relatively small percentages of respondents select- ing the two extreme response categories: the UN is With the collapsed categories illustrated in doing a very good or a very poor job. Furthermore, Table 14-8, we can now rather easily read across although it might be tempting to read only the sec- the several national percentages of people who said ond line of the table (those saying \"good job\"), that the UN was doing at least a good job. Now the would be impropeL Looking at only the second United States appears the most positive; Germany, row, we would conclude that West Germany and Britain, and France are only slightly less positive the United States were the most positive (46 per- and are nearly indistinguishable from one another; cent) about the UN's performance, followed closely and Japan stands alone in its quite low assessment by France (45 percent), with Britain (39 percent) of the UN's performance. Although the conclusions less positive than any of those three and Japan (11 to be drawn now do not differ radically from what percent) the least positive of all. we might have concluded from simply reading the second line of Table 14-7, we should note that This procedure is inappropriate in that it ig- Britain now appears relatively more supportive. nores all those respondents who gave the most pos- itive answer of all: \"very good job.\" In a situation Here's the risk I'd like to spare you. Suppose like this, you should combine or \"collapse\" the two you had hastily read the second row of Table 14-7 ends of the range of variation. In this instance, and noted that the British had a somewhat lower

418 Chapter 14: Quantitative Data Analysis TABLE 14-9 the \"Don't Knows\" West Germany Britain France Japan United States Good job or better 65% 55% 65% 20% 57% Poor job or worse 35% 45% 35% 81% 44% assessment of the job the UN was doing than was respondents said they didn't know. This means that true of people in the United States, West Germany, those who said \"good\" or \"bad\" job-taken to- and France. You might feel obliged to think up an gether-represent only 74 percent (100 minus 26) explanation for why that was so-possibly creating of the whole. If we divide the 48 percent saying an ingenious psychohistorical theory about the \"good job or better\" by 0.74 (the proportion giving painful decline of the once powerful and dignified any opinion), we can say that 65 percent \"of those British Empire. Then, once you had touted your ,\"ith an opinion\" said the UN was doing a good or \"theory\" about someone else might point out that very good job (48%10.74 = 65%). a proper reading of the data would show the British were actually not really less positive than the other Table 14-9 presents the whole table with the three nations. This is not a hypothetical risk. Errors \"don't knows\" excluded. Notice that these new data like these happen frequently, but they can be offer a somewhat different interpretation than the avoided by collapsing answer categories where previous tables do. Specifically, it would now ap- appropriate. pear that France and West Germany were the most positive in their assessments of the UN, \"'ith the Handling \"Don't Knows\" United States and Britain a bit lower. Although Japan still stands out as lowest in this regard, it has Tables 14-7 and 14-8 illustrate another common moved from 12 percent to 20 percent positive. problem in the analysis of survey data. It's usually a good idea to give people the option of saying \"don't At this point having seen three versions of the know\" or \"no opinion\" when asking for their opin- data, you may be asking yourself, Which is the ions on issues. But what do you do with those an- right one? The answer depends on your purpose in swers when you analyze the data? analyzing and interpreting the data. For example, if it's not essential for you to distinguish \"very good\" Notice there is a good deal of variation in the from \"good,\" it makes sense to combine them, be- national percentages saying \"don't know\" in this in- cause it's easier to read the table. stance, ranging from only 10 percent in the United States to 41 percent in Japan. The presence of sub- Whether to include or exclude the \"don't stantial percentages saying they don't know can knows\" is harder to decide in the abstract. It may confuse the results of tables like these . For example, be a very important finding that such a large per- was it simply because so many Japanese didn't ex- centage of the Japanese had no opinion-if you press any opinion that they seemed so much less wanted to find out whether people were familiar likely to say the UN was doing a good job? \"'ith the work of the UN, for example. On the other hand, if you wanted to know how people Here's an easy way to recalculate percentages, might vote on an issue, it might be more appropri- with the \"don't knows\" excluded. Look at the ate to exclude the \"don't knows\" on the assump- first column of percentages in Table 14-8: West tion that they wouldn't vote or that ultimately they Germany's answers to the question about the would be likely to dhide their votes between the UN's performance. Notice that 26 percent of the two sides of the issue. In any event, the [rutlz contained mthin your data is that a certain percentage said they didn't

Bivariate Analysis 419 know and the remainder divided their opinions in two sectors should exclude this clinic and whatever manner they did. Often, it's appropriate should only compare consultations taken by a to report your data in both forms-with and ,\"ith- single doctor in both sectors. This subsample of out the \"don't knows\"-so your readers can draw cases revealed that the difference in length be- their own conclusions. tween I\\T}{S and private consultations was now reduced to an average of under 3 minutes. This Numerical Descriptions was still statistically Significant, although the in Qualitative Research significance was reduced. Finally, however, if I compared only new patients seen by the same Although this chapter deals primarily ,vith quanti- doctor, I\\T}{S patients got 4 minutes more on tative research, the discussions also apply to quali- the average-34 minutes as against 30 minutes tative studies. Numerical testing can often verify in the private clinic. the findings of in-depth, qualitative studies. Thus, for example, when David Silverman wanted to (J 993. 163-64) compare the cancer treatments received by patients in private clinics with the cancer treatments in This example further demonstrates the special Britain's National Health Service, he primarily power that can be gained from a combination of chose in-depth analyses of the interactions be- approaches in social research. The combination of tween doctors and patients: qualitative and quantitative analyses can be espe- cially potent. My method of analysis was largely qualitative and ... I used extracts of what doctors and Bivariate Analysis patients had said as well as offering a brief ethnography of the setting and of certain be- In contrast to univariate analysis, subgroup com- havioural data. In addition, however, I con- parisons involve two variables. In this respect sub- structed a coding form which enabled me to group comparisons constitute a kind of bivariate collate a number of crude measures of doctor analysis-that is, the analysis of two variables si- and patient interactions. multaneously. However, as ,vith univariate analy- sis, the purpose of subgroup comparisons is largely (1993 163) descriptive Most bivariate analysis in social re- search adds another element: determining relation- Not only did the numerical data fine-tune Silver- ships between the variables themselves. Thus, uni- man's impressions based on his qualitative observa- variate analysis and subgroup comparisons focus tions, but his in-depth understanding of the situa- on describing the people (or other units of analysis) tion allowed him to craft an ever more appropriate under study, whereas bivariate analysis focuses on quantitative analysis. Listen to the interaction be- the variables and their empirical relationships. tween qualitative and quantitative approaches in this lengthy discussion: Table 14-10 could be regarded as an instance of subgroup comparison: It independently describes My overall impression was that private consul- the religious services attendance of men and tations lasted considerably longer than those held in the NHS clinics. When examined, the bivariate analysis The analysis of two variables si- data indeed did show that the former were al- multaneously, for the purpose of determinina the most t'ivice as long as the latter (20 minutes as against 11 minutes) and that the difference was empirical relationship between them. The co~stnlC­ statistically highly significant. However, I re- called that, for special reasons, one of the NHS lion of a simple percentage table or the computation clinics had abnormally short consultations. I of a simple correlation coefficient are examples of bi- felt a fairer comparison of consultations in the variate analyses

420 \" Chapter 14: Quantitative Data Analysis TABLE 14-10 for any given table. In Table 14-10, for example, I've divided the group of subjects into two sub- Church Attendance Reported by Men and Women in 2000 groups-men and women-and then described the behavior of each subgroup, That is the correct Weekly Men Women method for constructing this table. Notice, however, Less often that we could-however inappropriately-con- 25% 33% struct the table differently. We could first clivide the 100% = 75 66 subjects into clifferent degrees of religious services (1,199) (1,538) attendance and then describe each of those sub- groups in terms of the percentage of men and Note: Rounding to the nearest whole percentages may result in atotal of 99% or women in each. This method would make no sense 101% in some cases.This is referred to as a\"rounding error.\" in terms of explanation, however. Table 14-10 sug- gests that your gender will affect your frequency of women, as reported in the 2000 General Social religious services attendance. Had we used the Survey. It shows-comparatively and descrip- other method of construction, the table would sug- tively-that the women under study attended gest that your religious services attendance affects church more often than the men clid. However, the whether you're a man or a woman-whid1 same table, seen as an explanatory bivariate analy- makes no sense. Your behavior can't determine sis, tells a somewhat different story. It suggests that your gender. the variable gender has an effect on the variable church attendance. That is, we can view the behavior A related problem complicates the lives of new- as a dependent variable that is partially determined data analysts. How do you read a percentage table? by the independent variable, gender. There is a temptation to read Table 14-10 as fol- lows: \"Of the women, only 33 percent attended re- Explanatory bivariate analyses, then, involve ligious services weekly, and 66 percent said they at- the \"variable language\" introduced in Chapter 1. In tended less often; therefore, being a woman makes a subtle shift of focus, we're no longer talking about you less likely to attend religious services fre- men and women as different subgroups but about quently.\" This is, of course, an incorrect reading of gender as a variable: one that has an influence on the table. Any conclusion that gender-as a vari- other variables. The theoretical interpretation of able-has an effect on religious services attendance Table 14-10 might be taken from Charles Glock's must hinge on a comparison between men and Comfort Hypothesis as discussed in Chapter 2: women. Specifically, we compare the 33 percent with the 25 percent and note that women are more L Women are still treated as second-class citizens likely than men to attend religious services weekly. in U.S. society. The comparison of subgroups, then, is essential in reading an explanatory bivariate table. 2. People denied status gratification in the secular society may turn to religion as an alternative In constructing and presenting Table 14-10, I've source of status. used a convention called percentage down. This term means that you can add the percentages down each 3. Hence, women should be more religious column to total 100 percent (with the possibility than men. of a rounding error, as noted in the table). You read this form of table across a row, For the row The data presented in Table 14-10 confirm this rea- labeled \"weekly,\" what percentage of the men at- soning. Thirty-three percent of the women attend tend weekly? What percentage of the women religious services weekly, as compared with 25 per- attend weekly? cent of the men. The direction of percentaging in tables is arbi- Adding the logic of causal relationships among trary, and some researchers prefer to percentage variables has an important implication for the con- across. They would organize Table 14-10 so that struction and reacling of percentage tables. One of the chief bugaboos for new data analysts is decid- ing on the appropriate\"direction of percentaging\"

Bivariate Analysis\" 421 \"men\" and \"women\" were shown on the left side of TABLE 14-11 the table, identifying the two rows, and \"weekly\" and \"less often\" would appear at the top to identify Hypothetical Data Regarding Newspaper Editorials the columns. The actual numbers in the table on the Legalization of Marijuana would be moved around accordingly, and each row of percentages would total approximately 100 per- Editorial Policy Community Size cent. In that case, you would read the table down a Toward Legalizing column, still asking what percentage of men and Marlj'uana Under 100,000 Over 100,000 women attended frequently. The logic and the con- clusion would be the same in either case; only the Favorable 11% 32% form would cliffer. Neutral 29 40 Unfavorable 60 28 In reading a table that someone else has con- (127) (438) structed, therefore, you need to find out in which 100% = clirection it has been percentaged. Usually this will be labeled or be clear from the logic of the variables Table 14-11 presents some hypothetical data being analyzed. As a last resort, however, you describing the eclitorial policies of rural and urban should add the percentages in each column and newspapers. Note that the unit of analysis in this each row. If each of the columns totals 100 percent, example is the individual eclitoriaL Table 14-11 the table has been percentaged down. If the rows tells us that there were 127 eclitorials about mari- total 100 percent each, it has been percentaged juana in our sample of newspapers published in across. The rule, then, is as follows: communities with populations under 100,000. (Note that this cutting point is chosen for simplicity 1. If the table is percentaged down, read across. of illustration and does not mean that rural refers to a community of less than 100,000 in any absolute 2. If the table is percentaged across, read down. sense.) Of these, II percent (14 eclitorials) were fa- vorable toward legalization of marijuana, 29 per- Percentaging a Table cent were neutral, and 60 percent were unfavor- able. Of the 438 editorials that appeared in our Figure 14-5 on page 422 reviews the logic by sample of newspapers published in communities of which we create percentage tables from two vari- more than 100,000 residents, 32 percent (140 ecli- ables. I've used as variables gender and attitudes torials) were favorable toward legalizing marijuana, tOlVard eqllaliry for men and women. 40 percent were neutral, and 28 percent were unfavorable. Here's another example. Suppose we're inter- ested in learning something about newspaper edi- When we compare the editorial policies of rural torial policies regarcling the legalization of mari- and urban newspapers in our imaginary study, we juana. We undertake a content analysis of editorials find-as expected-that rural newspapers are less on this subject that have appeared during a given favorable toward the legalization of marijuana than year in a sample of daily newspapers aCTQSS the na- urban newspapers are. We determine this by not- tion. Each editorial has been classified as favorable, ing that a larger percentage (32 percent) of the ur- neutral, or unfavorable toward the legalization of ban editorials were favorable than the percentage marijuana. Perhaps we wish to examine the rela- of rural ones (II percent). We might note as well tionship between editorial policies and the types of that more rural than urban editorials were unfa- communities in which the newspapers are pub- vorable (60 percent compared with 28 percent). lished, thinking that rural newspapers might be Note that this table assumes that the size of a com- more conservative in this regard than urban ones. munity might affect its newspapers' editorial poli- Thus, each newspaper (hence, each eclitorial) has cies on this issue, rather than that editorial policy been classified in terms of the population of the might affect the size of communities. community in which it is published.

a. Some men and women who either favor (+) gender equality or don't (-) favor it b. Separate the men and the women (the independent variable) . c. Within each gender group, separate those who favor equality from those who don't (the dependent variable). d. Count the numbers in each cell of the table. e. What percentage of the women favor equality? f. What percentage of the men favor equality? 80% g. Conclusions. Favor equality - - - + - - - While a majority of both men and women favored gender equality, Don't favor equality ----11---- women were more likely than men to do so . Thus, gender appears to be one of the causes of attitudes toward Total 100% 100% sexual equality . FIGURE 14-5 Percentaging aTable

Bivariate Analysis 423 Constructing and Reading Tables such as the ones we've been examining Bivariate Tables are commonly called contingency tables: Values of the dependent variable are contingent on (de- Let's now review the steps involved in the con- pend on) values of the independent variable, Al- struction of explanatory bivariate tables: though contingency tables are common in social science, their format has never been standardized. L The cases are divided into groups according to As a result. you'll find a variety of formats in re- the attributes of the independent variable. search literature. As long as a table is easy to read and interpret. there's probably no reason to strive L Each of these subgroups is then described in for standardization. However. there are several terms of attributes of the dependent variable. guidelines that you should follow in the presenta- tion of most tabular data. 3. Finally, the table is read by comparing the inde- pendent variable subgroups with one another 1. A table should have a heading or a title that in terms of a given attribute of the dependent succinctly describes what is contained in the variable. table. Following these steps, let's repeat the analysis 2. The original content of the variables should be of gender and attitude on sexual equality. For the clearly presented-in the table itself if at all reasons outlined previously, gender is the indepen- possible or in the text with a paraphrase in the dent variable; attillide [award sexual equality consti- table. This information is especially critical tutes the dependent variable. Thus, we proceed as when a variable is derived from responses to an follows: attitudinal question, because the meaning of the responses will depend largely on the word- 1.. The cases are divided into men and women. ing of the question, 2 Each gender subgrouping is described in terms 3. The attributes of each variable should be clearly of approval or disapproval of sexual equality. indicated. Though complex categories will have 3. Men and women are compared in terms of the to be abbreviated, their meaning should be clear in the table and, of course, the full de- percentages approving of sexual equality. scription should be reported in the text. In the example of editorial policies regarding 4. When percentages are reported in the table, the legalization of marijuana, size ofc0Il1IJ111l1iry is the the base on which they are computed should independent variable, and a newspaper's editorial pol- be indicated. It's redundant to present all the icy the dependent variable. The table would be con- raw numbers for each category, because these structed as follows: could be reconstructed from the percentages and the bases. Moreover, the presentation of 1. Divide the editorials into subgroups according both numbers and percentages often confuses to the sizes of the communities in which the a table and makes it more difficult to read. newspapers are published. 5. If any cases are omitted from the table 2. Describe each subgroup of editorials in terms of because of missing data (\"no answer,\" for ex- the percentages favorable, neutraL or unfavor- ample), their numbers should be indicated in able toward the legalization of marijuana. the table. 3. Compare the two subgroups in terms of the contingency table A format for presenting the percentages favorable toward the legalization of relationships among variables as percentage marijuana. distribu tions, Bivariate analyses typically have an explana- tory causal purpose. These two hypothetical exam- ples have hinted at the nature of causation as social scientists use it

424 Chapter 14: Quantitative Data Analysis Introduction TABLE 14-12 to Multivariate Analysis Multivariate Relationship: Religious Services The logic of multivariate analysis, or the analysis Attendance, Gender, and Age of more than two variables simultaneously, can be seen as an extension of bivariate analysis. Speci- \"How often do you attend religious services?\" fically, we can construct multivariate tables on the basis of a more complicated subgroup description About weekly* Under 40 40 and Older by following essentially the same steps outlined for Less often Men Women Men Women bivariate tables. Instead of one independent vari- able and one dependent variable, however. we'll 100% = 20% 23% 30% 38% have more than one independent variable. Instead 80 77 70 62 of explaining the dependent variable on the basis of (504) (602) (695) (936) a single independent variable, we'll seek an expla- nation through the use of more than one indepen- 'About weekly = \"More than once aweek,\"\"Weekly,\" and \"Nearly every week\" dent variable. Soune: General Social Survey, 2000 Let's return to the example of religious services Among men, the respective figures are 20 and attendance. Suppose we believe that age would also 30 percent. affect such behavior (Glock's Comfort Hypothesis 2. Within each age group, women attend slightly suggests that older people are more religious than more frequently than men, Among those re- younger people). As the first step in table construc- spondents under 40,23 percent of the women tion, we would divide the total sample into sub- attend weekly, compared with 20 percent of groups based on the attributes of both independent the men. Among those 40 and over, 38 percent variables simultaneously: younger men, older men, of the women and 30 percent of the men at- younger women, and older women. Then the sev- tend weekly. eral subgroups would be described in terms of the 3. As measured in the table, age appears to have a dependent variable, religious services attendance, and greater effect on attendance at religious services comparisons would be made. Table 14-12, from an than gender does. analysis of the 1973-1993 General Social Survey data, is the result. 4. Age and gender have independent effects on religious service attendance. Within a given at- Table 14-12 has been percentaged down and tribute of one independent variable, different therefore should be read across. The interpretation attributes of the second still affect behaviors. of this table warrants several conclusions: 5. Similarly, the two independent variables have a 1. Among both men and women, older people cumulative effect on behaviors. Older women attend religious services more often than attend the most often (38 percent), and younger younger people do. Among women, 23 percent men attend the least often (20 percent). of those under 40 and 38 percent of those 40 and older attend religious services weekly. Before I conclude this section, it will be useful to note an alternative format for presenting such data. multivariate analysis The analysis of the simulta- Several of the tables presented in this chapter are neous relationships among several variables. Exam- somewhat inefficient. When the dependent vari- ining simultaneously the effects of age, gender, and able, religious attendance, is dichotomous (having ex- sodal class on religiosity would be an example of mul- actly two attributes), knowing one attribute permits tivariate analysis., the reader to reconstruct the other easily. Thus, if we know that 23 percent of the women under 40 attend religious services weekly, then we know automati- cally that 77 percent attend less often. So reporting the percent attending less often is unnecessary.