Cross-Cultural Sample (SCCS)
Philip Silverman and Jacquelyn Messinger
Introduction to the Module
The comparative method has been one of the hallmarks of anthropology since its inception in the middle of the 19th Century. Although at various times it was neglected as researchers focused on the less daunting task of understanding a single cultural system, it remains one of the most distinctive contributions that can be made by a science whose data base attempts to account for all cultural systems both temporally and spatially throughout the world. But the efforts of comparativists have been fraught with controversy and methodological barriers that can defeat the faint-hearted. Examination of the data made available here will provide students with an opportunity to evaluate the usefulness of testing hypotheses globally within the framework of pre-coded variables from a systematic sample of the known cultures in the world. In what follows the terms culture and society are used interchangeably.
It may be instructive for us to examine briefly the history of these cross-cultural (sometimes called holocultural) efforts. In the 1930s George Peter Murdock of Yale University revived an interest in large-scale comparisons by initiating the Cross-Cultural Survey, which was reorganized and expanded in the 1940s under the name of the Human Relations Area File (HRAF). This was basically an attempt to provide researchers with the data from a large number of societies, mostly data collected in the form of ethnographies, and so organized that comparisons across many cultures would be facilitated. This effort first required Murdock to make a listing of all known cultures (Murdock, 1975) before selecting those that provided the most detailed information (one of the criterion for inclusion in the file), and a list of categories into which the data could be organized. This later effort (Murdock, et. al., 1982) resulted in the creation of over 800 categories from which one could choose and then find in one place all the data from the original publications available in any language (all translated into English) relevant to that topic from a given society. At present the data from almost 400 cultures have been included in the HRAF.
All this was initiated prior to the technology associated with high speed computers. By the 1950s Murdock began a new effort that involved coding the material from the written sources into predetermined categories so that the data could be analyzed by the increasingly more efficient machine technology. He himself coded the ethnographic material on selected variables for over (believe it or not) one thousand cultures. Because the ethnographic sources are particularly rich on family relations, and also because Murdock had a special interest in the subject, many of the variables he coded had to do with details of the kinship system.
However, this initial effort by Murdock, which came to be known as the Ethnographic Atlas (Murdock, 1957), ignored the distribution of and relationship between the large number of cultures included in the Ethnographic Atlas. Since any sample of this magnitude invariably included cultures that were closely related historically, linguistically, and geographically, one could not assume independence of cases as required for statistical analysis. Consequently, Murdock began to organize the thousand plus societies into some 400 clusters of the most closely related ones, and then after creating more stringent criteria, reduced the clusters to 200 "sampling provinces" from which it would be possible to choose only one case for inclusion in a particular study. Thus all the societies included in a given sampling province shared, according to judgements Murdock made explicit, too many historical connections to assure independence of cases. By the late 1960s, and now working with Douglas White, Murdock eliminated several sampling provinces for which the data were inadequate, recombined several others that on closer examination appeared too similar, and ended up choosing one society from each of 186 sampling provinces to create the Standard Cross-Cultural Sample (Murdock and White, 1968). This represents a global stratified sample of both contemporary and historical as well as literate and preliterate societies.
Since the creation
of the SCCS Murdock and his associates, and a growing number of specialists
in various substantive areas of the study of culture, have been coding information
from the societies in this sample. Thus the data set we have available here
provides a large number of variables, although by no means all of them, on many
aspects of culture and several key aspects of the natural environment as well.
As you peruse the Codebook you will notice that for
a number of variables the data are missing from many of the societies. This
is partly due to the enormous variety of cultural elements found globally, but
also because of the obvious difference in interests that particular researchers
have had. Nevertheless, the large number of variables available provide opportunities
to combine a number of conceptually similar variables into higher order constructs.
More on this as we proceed. The following is a list of the basic cultural categories
of the over 500 variables included in this data set. For details on the variables
found within each of the categories, the Codebook should
Categories of Culture in SCCS Data Set
1. Subsistence Economy and Supportive Practices
2. Infancy and Early Childhood
3. Settlement Pattern and Community Organization
4. Political Organization
5. Division of Labor (by gender only)
6. Cultural Complexity
7. Sexual Attitude and Practice
8. Climate Data
9. Ethnographic Atlas (includes most of variables Murdock originally coded)
10. Traits Inculcated in Childhood
11. Agents and Techniques of Child Training
12. Adolescent Initiation Ceremonies
13. Relative Status of Women
14. Cultural Theories of Illness
15. Female Power and Male Dominance
16. Female Status
17. Quality Control Variables
18. Husband and Wife Relationships
19. Political Decision Making and Conflict
20. Adolescent Sexual Behavior
21. Polygyny: Form and Frequency
22. Nature of Warfare
One of the advantages of this sample is the data have been pinpointed in both time and space. This means that for each society the data are coded from one time period (i.e., Japan in the early 1950s, or Ireland in the 1930s), and from one locality within the society, most usually a specific community in which the research was conducted. Because of these necessary temporal and spatial boundaries, it is possible that certain cultural items exist in the society as a whole, but are not found in the particular community from which these data derive. This will be more likely to occur in the larger, more complex societies like China and India. Thus, you must keep in mind that the unit of analysis for these data is a community.
The number of cross-cultural
studies available using large-scale samples now numbers in the hundreds. In
addition, a fair proportion of these studies have used the SCCS since it is
arguably the most effective global sample of cultures. All of the variables
in this data set have been used in one or another of these studies so it is
possible to review how the data were handled by others. The most detailed review
of the variety of studies in this genre can be found in Levinson and Malone
(1980). Another useful source is Naroll (1983), with a focus on countries as
the unit of analysis as well as some studies using the cultural units in the
Theoretical disputes in anthropology at times have led to bitter acrimony and sometimes sadly even personal attacks on opponents. This is complicated by the fact that there is a welter of theories contending for attention, with often considerable overlapping concepts and explanations, all of which creates great confusion for the novice. However, for our purposes we can reduce this messy diversity to two basic paradigms that are dominant in the field today: materialism and idealism.
Materialism - the main premise is that culture above all else is a system by which humans adapt to their environment. It focuses on the material conditions of life, the practical problems that humans must face in their daily existence. This approach gives primary emphasis to the manner in which technology and economic relations determine other aspects of culture. A typical causal sequence used by materialists involves a three tier model: infrastructure (the techno-economic base) generates the structure (the social groupings that characterize a society), which in turn determines the superstructure (the ideology and expressive life of a people). Thus, this perspective has led researchers to show how varying ecological conditions relate to subsistence techniques, or how differing economic systems can predict family types and child training practices, which in turn lead to distinctive religious expressions or worldviews as part of the superstructure.
Idealism - the main premise of this paradigm is that culture is, above all, a symbolic system. The focus here is on how humans create meaning in their lives by the selective receiving, manipulating, and transmitting of symbols. The basic problem is to uncover the meaning behind social action which, it is insisted, can only be understood with reference to the specific contexts within which humans are engaged in ongoing negotiation. Unlike materialists who tend to focus on efficiency as the primary explanatory tool, symbolic theorists recognized a variety of values that can influence decision making, including power, prestige, intellectual appeal, even aesthetic attraction. This leaves much greater room for recognition of individual variation, thus allowing for psychological explanations as well as cultural ones.
Unlike the materialists
who embrace a methodology distinctive of the natural sciences, idealists are
much more eclectic, often eschewing the constraints imposed by scientific procedures
for a more fluid, humanistic approach. Because this perspective tends to be
more emic (the insiderís view) and ideographic (avoidance of generalizations
beyond a particular case), the value of cross-cultural data bases is not recognized
by many idealists. However, it is possible to find studies in which primacy
is given to ideology and basic values as the major explanation for other aspects
of culture, thus reversing the causal sequencing preferred by materialists.
In addition, there is a considerable number of studies in which child training
practices are shown to have psychological and cultural consequences and are
not necessarily inconsistent with the causal sequencing preferred by materialists.
Whereas Karl Marx is the patron saint of materialist explanation, idealists
seek sustenance from the theories of sociologist Max Weber and psychologist
Example 1: Bivariate Analysis with Ordinal Scale Variables
A major concern in anthropology is understanding the nature of both biological and cultural evolution. Materialism has been the most effective theoretical perspective in tracing the macro-changes associated with evolution. Harris (1979) has elaborated a framework for determining broad causal sequences that can be employed in testing propositions based on a division of cultural materials into three levels mentioned earlier: infrastructure, the techno-economic base; structure, the social and political groupings; and superstructure, the ideological and expressive dimensions including religion, worldview, art, games and sports. Following in a long line of theorists that goes back at least to Marx, Harris argues that infrastructure determines both the structure and the superstructure. This framework provides a number of testable hypotheses that can be examined with the SCCS data.
Our data set contains a number of variables that measure the complexity of various aspects of culture. These measures of complexity can be used as rough indices of evolutionary complexity based on the elaborateness of the technology. As with many of the variables in this data set, the cultural complexity ones are organized as an ordinal scale. We will first illustrate how a materialist theoretical perspective can be explored using the Crosstabs procedure in SPSS. Variable 153 is a measure of Technological Specialization based on an ordinal scale of five levels of complexity. The frequencies for this variable can be examined in Table 1. [How to create Frequency Distributions in SPSS]
We have slightly revised the value labels in these tables in order to fit them into the table format generated by SPSS. The different levels of complexity represent technological characteristics of a very general nature and can be found in a variety of distinctive cultural contexts as the society becomes more complex. We can now relate this to a structural variable, as defined by the materialist paradigm, to determine if there is a concomitant increase in complexity. We have chosen Variable 157 Political Integration. The frequencies for this variable can be examined in Table 2.
Again we have an
ordinal scale of five groups based on increasing levels of political organization
beyond the community level (such as county, state, and federal levels). Given
we have two ordinal scales, we can use Gamma as the measure of association to
determine how strong the relationship is between these two variables. Table
3 presents the crosstabulation with Technological Specialization treated as
the independent variable and organized into columns, and Political Integration
as the dependent variable.
[How to create Crosstabs in SPSS]
Table 3 indicates a very strong correlation between the two variables, a Gamma of over .60 and a Chi square of less than .00001. Quite clearly, as technological specialization increases there is an equivalent increase in political complexity. These results are hardly surprising given the obvious connection between economic and political factors. However, less obvious is the impact of the infrastructure or structure on the superstructure. To examine this we have chosen a religious variable, Variable 238 High Gods, and hypothesize that higher levels of political complexity will predict the increasing involvement of God in human affairs, as conceptualized by a particular culture (this hypothesis is based on the work of Swanson, 1964). First we will examine the High Gods variables; the frequencies can be found in Table 4.
In Table 4 the values relating to "active" indicate whether or not God is actively involved in human affairs, and the values relating to "moral" indicate whether or not God is also supportive of human morality. Thus the variable can be considered an ordinal scale of increasing involvement of God in human affairs. Notice that there are 18 societies for which there is no information on the concept of God. Where data are absent and thus coded as "Missing," those cases will be excluded from any statistical procedures. Thus, in the following table, only 168 societies (186 is the total number of societies in the sample, minus 18 leaves 168 usable cases) are included in the calculations. Notice in Table 4 that the "Valid Percent" column gives the percentage distribution of the cases excluding those that are missing.
We can now relate this religious conception to political complexity. Table 5 presents the results. For this crosstabulation we have arranged the Political Integration variable in columns because here it is the independent variable and High Gods is the dependent variable. By so doing we are conforming to the causal sequencing predicted by materialist theory.
We can now examine the column percentages in the cells of Table 5. If the hypothesis is confirmed, then the greater proportion of cases should be in the upper left and lower right quadrants. In general, inspection of the cells supports this conclusion, although the distribution is by no means perfect. Indeed, the relationship is not as strong as the previous crosstabulation in Table 3, as Gamma equals .41, but it remains substantial and is also significant at slightly above the .001 level. Thus, we have shown a correlation exists between the three levels of culture - infrastructure, structure, and superstructure - as conceived by the materialist theoretical perspective.
Before leaving Table 5 it is useful to examine more closely the so-called "negative cases;" i.e., those cases that do not conform to the hypothesis. For example, in the uppermost right hand cell one finds 8 societies which are at the highest level of political complexity, but the belief in High Gods is absent. It is possible to determine which societies fall into this cell by listing the cases (there is a LIST CASES function in SPSS), and then by examining which societies these are, attempt to find an explanation for this seeming anomaly. With respect to the negative cases in Table 5, it might well be that these are societies which share a particular religious belief system, perhaps Buddhism, where political complexity existed in the absence of a belief in a High God. Presumably, such societies find alternative mechanisms for inculcating a respect for authority without invoking the spiritual power of a supreme deity. Alternatively, or perhaps also in conjunction with Buddhism, these are societies with an elaborate belief in ancestor worship whereby ancestors are actively involved in the affairs of their living descendents rather than High Gods, as was found in most other societies.
Attempt to look
for other variables in the data set that are part of the superstructure or structure
as conceived by materialists, and determine if they are related to increasing
levels of infrastructural complexity. Keep in mind the range of cultural behavior
that is considered part of the superstructure. This might require a careful
perusal of all the variables in the Codebook. Any support
you find for the way in which the superstructure responds to changes in either
the structure or infrastructure potentially could add to our understanding of
Example 2: Bivariate Analysis with an Interval Scale Variable
In our previous example we crosstabulated two ordinal scale variables. Somewhat more powerful procedures are available if the data are organized on an interval or ratio scale level. Continuing with the materialists paradigm we can examine how ecological conditions constrain various elements of culture. There is an obvious link between ecological conditions and the subsistence techniques available to particular populations. The Innuit (Eskimos) have little choice but to pursue a hunting and gathering economy in the sub-Arctic environment of Alaska and northern Canada. Other settings allow for greater flexibility to be sure, but there is typically a close correlation between the physical environment and economic possibilities. Less obvious, however, is how ecological conditions may also influence aspects of the structure or superstructure. Cultural values associated with sexuality, for example, vary widely throughout the world. Previous studies have shown that greater cultural complexity is inversely related to sexual permissiveness (Hobhouse, et. al, 1915; Cohen, 1969).
However, it is possible that the relationship between complexity and sexual freedom is more indirect in that complex societies tend to emerge under certain ecological conditions, and it is these conditions that impact sexual mores. We will argue that in the equatorial zones, areas of relatively high annual temperatures, where there is only a minimal need to insulate the body with clothing, and where physical structures provide little opportunity for privacy (see Maxwell, 1967), there is more likely to be a concomitant openness regarding sexual expression. Conversely, where the climate demands increasing insulation of the body by means of clothing, and where the physical structures in which people live provide greater privacy and thus allow for a greater range of "backstage" behavior, there will be greater restrictions imposed on sexual expression. If this can be demonstrated we will have linked together elements of the infrastructure (clothing and the structure of house types) with an aspect of the superstructure (values with respect to sexual expression).
Our data set provides a number of variables to test this hypothesis. First, we will examine a relevant ecological variable, Var 186, Mean Annual Temperature (C); the "C" indicates that the temperature is measured by the centigrade scale. The frequencies for this variable, as shown in Table 6, indicate that annual temperatures vary from a mean of -16 C to that of +29, as one could expect, an impressive range throughout the world. This ecological factor will represent the independent variable.
Now we can examine a dependent variable for which Var 165 has been chosen, Premarital Sex Attitudes-Female. We have chosen the female instead of the equivalent variable for males because, in general, the latter are allowed greater sexual freedom even in societies that impose sanctions on premarital sexual activity.
Table 7 indicates that there are six levels of tolerance coded for the 130 societies for which there are data available. Unfortunately, in order to use this variable the sample will be reduced by 56 cases, but that should still be a sufficient number of cases with which to work. Since the independent variable is an interval scale, we can use a procedure that can compare means, the T test. In order to compare differences in the means between two sub-sets in the sample, it is necessary to dichotomize Variable 165, the dependent variable. Thus, we want to determine if the mean annual temperature is significantly higher in those societies which allow greater freedom for premarital sex compared to those which donít. If one examines Table 7, it seems reasonable to choose a cutting point between values 2 and 3, so that one group will consist of all societies that either tolerate or expect premarital sex and the other group will consist of all societies that disapprove of this kind of activity to varying degrees. Using the t-test procedure we can determine if there is a statistically significant difference between the two groups of societies as just defined. Table 8 provides the results of running premarital sex attitudes against mean annual temperature. [How to perform T-tests in SPSS]
As Table 8 indicates the difference in means between the two groups is hardly discernible with less than a one degree Centigrade difference between the tolerant and the less tolerant societies. The tolerant group mean is 18.96 C and the less tolerant group mean is 18.79). As can be expected with such a minimal difference in means, it is not statistically significant, as indicated by the 2-tail significance coefficient in the equal row (P = .925). Thus, we cannot reject the null hypothesis.
Initial negative results should not lead one to abandon a hypothesis without putting up a fight. Perhaps the variable doesnít measure exactly what one wishes. There may be alternative ways of measuring what one is interested in. In this case, it might well be that mean annual temperature includes a range of climate types that vitiate the hypothesized relationship. For example, it would surely include desert regions which have a very high mean annual temperature, but where considerable protective clothing is worn. In fact, there are a number of pastoral societies living in desert climates who fit this description, and in addition, tend to have rigid norms regarding female premarital sexual activity associated with the male dominant ethic characteristic of nomadic pastoralists. These factors may be distorting the relationship that in actuality is there.
There is another ecological variable, however, that will allow us to control for this desert environment factor. VAR 189 Mean Annual Precipitation provides another interval scale based on the amount of rain that falls on average each year as measured by millimeters. By substituting this variable, we are able to eliminate from the higher range of precipitation the desert dwelling cultures, but still include most tropical societies where rainfall tends to be much higher than in other regions of the globe. We will not include a frequency table for VAR 189 because the range in precipitation is too great in that it is measured in such minute quantities (it varies from 1 millimeter per annum to 2,720 millimeters per annum). Instead, using the same criterion for dichotomizing VAR 165, Premarital Sex Attitudes for Females, Table 9 runs these two groups of societies against precipitation, VAR 189.
We now find a much more dramatic difference between the two groups of our sample. The more tolerant group has a mean annual precipitation of over 1529 millimeters, whereas the less tolerant group is a less than 1033 millimeters, a difference of almost 500 millimeters between the two groups. This time, reading from the Unequal row, since there is a significant difference based the Leveneís test of equality (P = .047), we find the difference in the two means is highly significant (P = .007). Thus, we are able to show that climatic factors can have a discernible impact on something as seemingly remote as premarital sexual attitudes.
Using the same or any of the other ecological variables available in the Codebook, attempt to show how environmental factors may be related to family structure, political organization, or the nature of relations between women and men in the society. Check out the variables that relate to these aspects of culture and attempt to formulate a reasonable hypothesis. Potential hypotheses should be discussed and evaluated in small group discussions before attempts at testing proceed.
Example 3: Multivariate Analysis using Composite Variables
This example will employ the perspective of idealism (discussed under the Theoretical Consideration section) to explore several relationships that have been of concern in the field of feminist theory. One of the most prevalent aspects of feminist theory is the recognition of womenís oppression (Sanday, 1981; Moore, 1988; and Fonow and Cook, 1991). Laws and taboos that restrict females and demean their status abound on a global scale. Most cultures not only have distinct strictures that are legally enforced, but in addition there are also norms of behavior that are considered acceptable, in fact, necessary, in order to be deemed "normal". In many cases this "expected" behavior is not only sexually repressive but it is based on, and enforced with, violence. Indeed, several researchers have claimed that where this form of violence is most severe, it tends to be associated with cultural ideologies that favor males and emphasize the concept of masculinity and male dominance.
For example, DeMeo utilized a global geographic review to empirically study human behavior and social institutions and found a very clear pattern. There was a correlation in pre-industrial tribal societies between a harsh environment, the rigid social and sexual subordination of women, the equation of masculinity with toughness and warlikeness, and the repression and/or distortion of sexual pleasure (cited in Eisler 1995:92). Furthermore, Harris (1989) has argued that womenís status in chiefdoms, which are societies that are intermediate in terms of elaborateness of the political organization, is directly related to the extent that men used their height and weight advantage to gain control over the technological processes necessary for production and warfare. A womanís inability to function as effectively as a man in this respect lowers her status in the society. Harris further argues that, from an evolutionist perspective, male specialization relative to warfare led to further specialization that "cumulatively point to a plausible explanation of almost every feature of the depressed status of women . . .in pre-industrial societies with similar forms of agriculture" (1989:328).
Both DeMeo and
Harris clearly are coming from a materialist perspective. However, the approach
taken here will focus on a mental factor as the independent variable and relate
it to aspects of the social structure, thus making use of a perspective more
typical of idealists in that it reverses the causal sequence of the materialists.
As you will see, aspects of the infrastructure will not be ignored, but they
will play a very different role within the approach that follows. Thus, we argue
that when a culture is characterized by an "ideology of male superiority," a
psychological mechanism is likely to emerge that we refer to as gender identity
insecurity, and this in turn leads to the abuse of women or the need to impose
severe control over their behavior. Where men are insecure about living up to
the male ethos of superiority over women, especially where their role as protector
or provider is given exaggerated emphasis, it often leads to severe control
over and/or violence towards women in the ingroup. With our data it is possible
either to support or discredit this argument.
Creating Composite Variables
In order to test this hypothesis it will be necessary to construct composite variables by combining several of the available ones in our data set. This is done by combining the codes from a number of different variables all of which must reflect the same general concept that one is attempting to measure. One must be able to justify the inclusion of all the variables that are being added together so that each is a specific index of the more general concept. Furthermore, one must ensure that all the variables are being added in the same direction; i.e., the inclusion of each represents either more or less of the variable to be created. For this module we are interested in creating several composite variables using the Compute procedure in SPSS. [How to create composite variables with the Compute procedure in SPSS].
We first construct the independent variable, Ideology of Male Superiority. Since the variable we are interested in can be indexed by a number of different ideological items, we must take account of as many of these specific beliefs as our data permit. Thus, Ideology of Male Superiority is comprised of the following variables (you can check the specific coding for these variables in the Codebook: Var 664: Ideology of male toughness, Var 621: Explicit view that men should and do dominate their wives, Var 625: High value placed on males being aggressive, strong, and sexually potent, Var 626: Belief that women are generally inferior to men, and Var 615: Wife to husband institutionalized deference. All the variables were measured on an ordinal scale from 1 to 3 with 1 being the lowest level of control. The resulting combined ordinal variable measures the level of ideology on a scale from 1 to 15, with 1 being the lowest measure and 15 the highest. This scale was then recoded into 3 levels by dividing the scale into three equal parts between 1 and 15. Thus, the original three level ordinal scale utilized in the original variables, 1=low, 2=medium, 3=high, was re-created for the composite variable of Ideology of Male Superiority.
The data in Table 10 show the frequency distribution for the summary variable Ideology of Male Superiority. The larger percentages of valid cases, almost 50%, are in the low strength category for this variable. The high level has only 17% of valid responses. Notice also that there are a relatively large number of missing cases, 43. Data on the ideological issues of concern here cannot always be found in ethnographies. The more arcane the subject matter, the less likely one can expect to find the necessary information. We will nevertheless press forward with the number of cases available to us, 143, a number that remains respectable.
Next we will construct the two dependent variables by the same composite variable procedure used for the independent variable. In both cases these are behavioral variables that are the consequence of the ideological of male superiority. The first, Control of Female Sexual Activity, is computed from four variables in the data set: Var 165: Premarital sex attitudes-female; Var167: Frequency of premarital sex-female; Var169: Extramarital sex; and Var 634: Control of sex. All the variables except for Var 634 were measured on an ordinal scale from 1 to 6 with 1 being the lowest level of control. Var 634, Control of sex, was measured the same but 6 was the lowest level and 1 the highest. It was recoded to reverse the order of values to maintain consistency with the other variables and allow for adding them together. As with the first composite variable, the Compute procedure was used in SPSS and the resulting scale was then divided into thirds to obtain measures representative of low, medium, and high levels of control. [How to Recode variables in SPSS.]
Table 11 presents the frequencies for Control of Female Sexual Activity. The largest percentage of responses fall into the medium level of control, with 42%, followed by low levels with 37%. As with the composite variable for male ideology, the lowest percentage of responses falls into the high level. With this variable we lost only 16 cases for lack of data, leaving 170 valid cases.
Condoned Violence Against Women is the second dependent variable. It was constructed from Var 667: Rape: incidents, reports, or thought of as a means of punishment for women, or as part of a ceremony; Var 754: Wife beating; and Var 620: Physical punishment of the spouse condoned. These variables were originally coded as dichotomies with either the behavior being present (1) or absent (2). Var 620, Physical punishment of spouse, however, was originally coded in three levels: only husband hitting wife -39 cases, physical punishment by neither - 16 cases, or either may hit the other with only 8 cases. Both hitting categories were coded together due to the small number of valid cases in the either category. This created a dichotomy measured as present or absent in order to facilitate computation with the other two variables. The resultant variable was an ordinal measure from 1 to 6, with 6 being the strongest level of condoned physical violence. For purposes of continuity the scale was broken down and recoded into three levels - low, medium, and high, creating a tripartite ordinal measure.
Table 12 presents the frequency distribution for the variable Condoned Physical Violence Against Women. As with the other variables, the larger percentage of valid cases fall into the low level of condoned violence. However, unlike the other composite variables, the high category has the next higher percentage with 30%.
This completes the construction of the composite variables necessary for this analysis. We can now proceed to the correlation of the variables, and the testing of the hypothesis.
To test the hypothesis that an ideology of male superiority can predict behaviors that sanction controlling female sexuality and condoning violence against females, we use the Correlation procedure in SPSS [How to determine Correlations in SPSS.]. This allows for the use of either parametric or non-parametric correlation procedures. Since our variables are scales at the ordinal level, a non-parametric statistic is chosen, Spearmanís Rho, one of the earliest measures used for ordinal-ranked data. It provides a correlation coefficient from Ė1 (highest negative correlation) to +1 (highest positive correlation) which indicates the degree of agreement or disagreement between two sets of ordinal scales. As the coefficient approaches 0, there is a lack of any correlation between the two.
The information in Table 13 presents the correlations between the variables. Both dependent variables are correlated with an Ideology of Male Superiority at levels that are statistically significant. The relationship to Control over Female Sexual Activity is somewhat weaker with a correlation coefficient just over .19, but still statistically significant at the .05 level. Somewhat more convincing is the relationship with Condoned Violence Against Females with a correlation coefficient of just under .30, and significant at the .01 level. Let us now pursue this analysis by incorporating an additional variable that could explain a bit further the correlations that were found. [How to determine Correlations in SPSS.]
Adding a Third Variable
Although we did find a relationship between an Ideology of Male Superiority and the two dependent variables, the moderate strength of the coefficients might suggest that there are other factors that also are involved as possible intervening variables. Earlier we suggested that this ideology sets up a situation of insecurity on the part of males which leads to ingroup hostility against women. It is also conceivable that this same gender insecurity could also be deflected to an outgroup. Such projection of internal conflicts corresponds to the notion of defense mechanisms as formulated by psychoanalytic theory. With this extension of the argument, we can now hypothesize that where the hostility engendered by an Ideology of Male Superiority is deflected to an outgroup, so that males are engaged in extensive warfare with neighboring groups, there is less likely to exist hostility towards women in the ingroup. Of course, it is conceivable that the hostility could be directed in both directions (ingroup and outgroup) as well. Whichever might be the case, we can test this through a re-examination of the previous correlation by controlling for the presence of warfare.
Several variables in our data set relate to the presence of outgroup hostility, including Var 679: Warfare or Fighting, Var 693: Frequency of Intercommunity Armed Conflict, and Var 715: Systematic Absences of Married Males (Military service, Labor elsewhere, Extended trade expeditions, etc.). While variables Var 679 and Var 693 deal exclusively with war and conflict, Var 715 encompasses considerably more instances where males may be absent than exclusively for warfare purposes. Of these variables the best fit for our purposes is Var 679, which is coded as a dichotomy: the lower level, warfare is absent or occasional/periodical, is coded 1; the higher level, warfare is frequent or endemic, is coded 2. Table 14 shows the frequency distribution for variable 679: Warfare or Fighting. As you can see, there is a considerable difference with 70% of the valid cases falling in the frequent or endemic category.
Given that this variable is a dichotomy, we can utilize a simple multivariate procedure in Crosstabs, whereby we can examine the original relationship as it is affected by each of the two values of Var 679. In using the Crosstabs program by controlling for a third variable, first enter the independent variable, Ideology of Male Superiority, in the columns section. Then enter the summary variable, Condoned Physical Violence Against Females, in the rows section. Finally, enter the control variable, Var 679: Warfare or Fighting, into the section marked "layer 1 of 1". This will break the crosstabs procedure into separate tables for each category of the intervening (control) variable. Go to the cells section and figure the percentages in the direction of the independent variable, in this case, the columns for Ideology of Male Superiority. This will result in percentage totals for each level of physical violence for the three categories of Ideology of Male Superiority. Next go to the statistics section. Check the boxes for Chi-square statistics and Gamma. We use Gamma for the correlation coefficient because the variables are ordinal measures. Click continue until you are back to the cross-tables window where you will click OK to generate the table and associated statistics. Follow the same procedures to create the tables for Control of Female Sexual Activity. [How to create crostabs controlling for a third variable in SPSS.]
Table 15 presents the data for the first of the two dependent variables, Condoned Physical Violence Against Women, controlling for the presence of absence of Warfare or Fighting. Notice that the upper section of Table 15 provides data for the relationship only for those cases where warfare is absent or only occasional, and the lower section for those cases where warfare is frequent or endemic. Because our sample has been divided by the control variable, so that, for example, in the upper section there are only 30 cases available for analysis, we have too many cells with less than five cases for the Chi square to be meaningful, but let us proceed with the analysis despite this limitation. The results of Table 15, although only suggestive, support the hypothesis that where aggression is directed to an outgroup, there is lesser likelihood that violence against women is condoned; i.e., where warfare is absent, the Gamma is much stronger, .63, than where warfare is present, Gamma = .30. Thus, the presence of outgroup hostility appears to mitigate physical violence toward females of the ingroup.
We now turn to the second of the dependent variables, Control of Female Sexual Activity. Table 16 presents the data for this correlation controlling again for the presence or absence of Warfare. Again we find that where Warfare is frequent or endemic, the Gamma is weaker (Gamma = .33), and where Warfare is absent, it is stronger (Gamma = .49). Thus, despite a reduced sample because of the absence of data in almost half of the societies, thus making it difficult to establish statistically significant measures, we also have found suggestive support that the presence of outgroup hostility as measure by the presence of Warfare appears to mitigate the relationship of an Ideology of male superiority with the second dependent variable, the Control over Female Sexual Activity.
Returning to the theoretical argument, we have to some extent reversed the materialistic argument put forward most persuasively by Harris (1989). Instead of warfare leading to a situation where the status of women is deflated, we began with an ideological argument, belief in male superiority, and demonstrated how the presence of warfare actually made it less likely, although clearly not impossible, that the position of women would be made more intolerable, as measured by violence against them and control over their sexuality.
Using the same
procedures utilized in the analysis performed above, pick several related variables
and create one or more composite variables to measure other aspects of human
behavior. Then choose a third variable to act as an intervening, or control,
variable. Try to demonstrate that the addition of this additional factor can
clarify or better explain a relationship that you have hypothesized between
Cohen, Yehudi A.