2.1. Study 1: folktale protocol

Folktale collections were chosen so as to maximize the geographical variability of the sample as well as variability in levels of cultural complexity. Specialized collections focusing on specific themes, plots, or character types (e.g., Hopi Trickster Tales, Hero Tales of the South Slavs) were rejected in favor of generic collections (Hopi Folktales, Traditional Tales of the South Slavs). All collections were of traditional tales, originally transmitted through the oral tradition. In all, the study includes tales from 48 different culture areas from all inhabited continents, varying widely in ecology, geographic location, racial and ethnic composition, political systems, religious beliefs, and levels of cultural complexity. All non-English tales had been translated into English, and the sample ran the gamut from polished fairy tales to literal transcriptions of tales told in traditional contexts. A list of the collections and tales coded can be accessed via the Internet at the following URL: http://www.science.mcmaster.ca/psychology/ehb/Gottschall(FolkTales).pdf.

We chose the folktale genre for our cross-cultural sample for several reasons. First, folktales are relatively short and allow for the processing of a large sample in a relatively brief span of time. Second, we hoped to, insofar as possible, diminish problems associated with cross-cultural borrowing and contamination. While the folktales in our sample were mostly compiled, edited, and translated by Westerners (and thus may reflect Western biases), the tales were originally composed and circulated in traditional societies and can be cautiously assumed to reflect the traditional attitudes and life ways of the populations that produced them. For the purposes of this study, we assume psychological equivalence between literary characters and real human beings. In fact, the relationship between the representation of behavior and psychology in literary works and behavior and psychology in real life is a subject that awaits systematic exploration. But, for the present, it seems safe to provisionally adopt the traditional assumption of literary scholars that literary works reflect both general aspects of human nature and particular aspects of given cultures, despite the fact that they also reflect the personal idiosyncrasies of writers and tellers (for discussion of this issue, see Carroll, in press).

Once suitable collections of tales were identified, each data collector (10 female and 5 male undergraduates at St. Lawrence University) was assigned a random selection of tales from each of three culture areas. (Several researchers were responsible for tales from four culture areas. They volunteered to code collections of tales that only arrived through interlibrary loan after the main portion of the study had been completed.) The coders were participants in a seminar on research methods, focusing on content analysis methodology in literary studies. The coders were made aware that controversies existed concerning constructivist versus evolutionary accounts of human mate preferences and were told that the results of their research—whether the results predominantly suggested cross-cultural regularity or extreme cross-cultural flux—would help to resolve these controversies.

Information was gathered on the main male and female protagonists and antagonists in each tale. For the purposes of our study, a main protagonist was defined as a character who plays a central role in the action and for whom the audience is led to root predominantly for rather than predominantly against. A main antagonist was defined as a character who plays a central role in the action, who acts as an obstacle to the goals of the protagonist(s), and for whom the audience is led to root predominantly against rather than predominantly for. A maximum of one coding form was filled out for each main male and female protagonist and antagonist per story. In the event that there were, for instance, two main female protagonists, coders focused only on the female protagonist who was judged most important and prominent. Thus, between one and four coding forms were completed for each work in the sample.

For all main characters, coders answered the following question: What single feature seems most important to the character in assessing the desirability of an existing or potential mate: (1) kindness, (2) possession of wealth and/or other material resources, (3) high social status, (4) physical attractiveness, (5) other, (6) impossible to answer? In the reporting of the results that follows, however, categories two and three are collapsed into one category. In all, information on mate preferences was available for 524 folktale characters, 246 male and 278 female. These figures do not include the relatively rare instances (8% of the characters in the two studies) where coders entered “other” for a character's mate preference.

Data analysis is reported for the sample as a whole and for each of six geographical regions: the circum-Mediterranean, East Eurasia, North America, South America, Africa, and the Insular Pacific combined with Southeast Asia and the Pacific Rim (see Appendix). Where possible our division of culture areas into geographical regions follows Murdock, 1957, Murdock, 1981, although it diverges in two significant ways. First, the analysis makes up for a shortfall of tales in Murdock's Insular Pacific region and a glut of tales in his East Eurasian region by grouping several culture areas in Southeast Asia and the Pacific Rim with tales from the Insular Pacific. Second, East African tales were grouped with Africa rather than with circum-Mediterranean tales. These divergences from Murdock result from the desirability of establishing regional samples of roughly equivalent size. The sample was also divided into two broad levels of cultural complexity. The first level consists of tales that circulated primarily in unassimilated band and tribal societies, though the tales may have only been written down after assimilation. The second level consists of tales that, while they may have originated in nonstate societies, circulated for long periods in preindustrial state societies. Because the line between these categories can obviously be fine, tales from culture areas that could not be confidently placed in one of these two categories were excluded from the analysis (see Appendix for description of the two subsamples).

2.2. Study 2: Western literature protocol

Study 2 was conducted in a different semester with different student coders (7 male and 5 female), but coding procedures were identical to Study 1. Our sources of data for Western literature were a random selection of Masterplots (1977) plot summaries; the summaries were read conjointly with complementary character summaries from the Cyclopedia of Literary Characters (1995). Masterplots is a collection of roughly 1200–1500 word plot summaries of the most prominent works in Western literature. The Masterplots summaries were complemented with character summaries from the Cyclopedia of Literary Characters. The Cyclopedia provided more in-depth information on individual characters with longer summaries (150–200 words) for the main characters. Thus, for every work of Western literature coders were in possession of a Masterplots summary as well as complementary information from the Cyclopedia of Literary Characters. Each coder was responsible for 20 randomly selected sets of plot/character summaries. While Masterplots includes a small percentage of nonfictional works (e.g., summaries of historical and philosophical works), all nonfiction was excluded from our analysis. From the 240 sets of summaries, information was available for the mate preferences of 309 characters, 188 male and 121 female. A list of all characters and works coded can be found at the URL address: http://www.science.mcmaster.ca/psychology/ehb/Gottschall(EurLit).pdf.

We chose to focus on summaries of Western literature rather than the works themselves for several reasons. First, while the constitutions of literary canons are in constant flux and highly controverted, Masterplots and the Cyclopedia represent something resembling a consensus on what works form the core of the Western literary tradition. Second, and most importantly, using summaries allowed us to derive basic information from a large number of literary works in a reasonable amount of time. Of course, using summaries also comes with costs. Foremost among these is the loss of valuable information that occurs whenever a long, complex literary work is reduced to a 1500-word gloss, or a complicated character is reduced to a 200-word summary. It is doubtlessly true that many of the summaries we read omitted details relevant to the mate preferences of characters. It is also doubtlessly true that information in the summaries was colored by the personal attitudes and prejudices of their authors, despite attempts to maintain objectivity. However, it should be noted that the Masterplots and Cyclopedia summaries are not the idiosyncratic products of one individual's vision: The choice of works in these references was determined collaboratively and the actual summaries were contributed by scores of English faculty from dozens of universities (see Masterplots, v–vii). In sharing the substantial task of choosing the works and authoring the summaries, the editors of these references guarded against the possibility that the biases of individual authors would compromise the reliability of the references as a whole.

2.3. Assessing intercoder reliability

As the wording of the coding question makes clear, coders were asked to make subjective judgments about which mate preference trait seemed most important to the characters being coded. Coder subjectivity bias is the main methodological obstacle in human coder content analysis. To diminish the scope for subjectivity bias we adopted some standard safeguards: We developed a coding question that was as uncomplicated as possible, we devised simple coding instructions, and we prepared a coding dictionary with simple definitions of all potentially ambiguous terms. For the four traits coded there were no significant differences between the codings produced by male and female coders.

Intercoder reliability was also assessed in two formal tests. Test 1 assessed the reliability of the folktale coders (Study 1) in coding an assortment of 11 culturally diverse folktales; intercoder reliability was 88%. Test 2 assessed the reliability of the Western literature coders (Study 2) in coding five randomly selected sets of plot and character summaries from Masterplots and the Cyclopedia, respectively; intercoder reliability was 74%. This latter result is at the lower end of the spectrum of an acceptable estimate of reliability for content analysis and therefore results from the Western literature study must be interpreted cautiously. Many content analysis practitioners strive for reliability rates of 80% or better and consider 70% to be the minimum level of adequacy Krippendorff, 1980, Neuendorf, 2002, Weber, 1990. Establishing reliability ratings prior to actual coding of the texts, as was done in the present study, rather than having multiple coders read and code all or some fraction of the different works, is well established in content analysis research. The advantage of this approach is that it facilitates the compilation of a large data set, though it does so at the cost of some degree of precision in assessing intercoder reliability for the coded data (for discussion of different methods of reliability testing see Krippendorff, 1980, Neuendorf, 2002, Weber, 1990).