While you have your own personal opinions about some things, you share attitudes with people around you about a lot of things. That's the notion of culture - shared meanings and feelings.

Variations Across Cultures

Sharing sentiments with others in your society is one side of culture. The other side is having different sentiments than people in other societies.

The chart below shows cultural sentiments for father, mother, and child as measured among people in the U.S.A., Canada, Japan, China, Germany, and Northern Ireland. This chart is based on female sentiments, but it would look about the same were male measurements used instead.

You can see that people in all six cultures agree that fathers, mothers, and children are not bad, and they agree that parents are powerful and children are powerless. However, aside from these generalities, major differences arise.

These results typify cross-cultural variations in sentiments. People in different cultures share some general perspectives, yet specific patterns of sentiments vary from one culture to another.

Cross-Cultural EPA Correlations

Hold on! What does it mean to say, Perspectives are shared, yet patterns vary? Are sentiments shared or do they vary across cultures?

Correlation analysis provides a way to assess the degree of cross-cultural sharing versus varying. It works like this: you try to predict sentiments in one culture from sentiments in another culture, and you compute a correlation coefficient to tell you how well you're doing. If you predict well, then sentiments are shared across the cultures; if you predict poorly, then sentiments vary across the cultures.

For example, matching U.S.A. and Canadian identities item by item can yield a formula for predicting evaluation ratings of Canadian identities from evaluation ratings of U.S.A. identities. Correlation coefficients measure how accurate such predictions are - perfect prediction corresponds to a coefficient of 1.00, and not being able to predict at all corresponds to a coefficient of 0.00.

Here is a graph showing how accurately other cultures' EPA sentiments about identities can be predicted from U.S.A. values. To illustrate, the front-most bar shows that Chinese females' evaluation ratings of identities can be predicted well from U.S.A. evaluation ratings of identities, the correlation coefficient being about 0.9. However, Chinese females' activity ratings of identities are predicted relatively poorly by U.S.A. ratings, with a correlation coefficient around 0.5.

Overall, the graph shows that the U.S.A. and Canada are rather similar in sentiments about identities, but patterns of sentiments about identities in other cultures diverge in various ways from U.S.A. sentiment patterns.

Sentiments about behaviors are more variable cross-culturally than sentiments about identities. This can be seen in the next graph, to be read like the last one.

U.S.A. evaluation ratings of behaviors predict evaluation ratings in other cultures well, but U.S.A. potency ratings of behaviors are only mediocre predictors of potency ratings in other cultures. U.S.A. activity ratings of behaviors are moderately good predictors in Canada, Germany, and Japan, but mediocre in Ireland, and completely useless for predicting Chinese activity ratings of behaviors.

The following table presents the numbers used in the last two graphs, as well as additional information on how well sentiments in each culture predict sentiments in other cultures. 

Studying the table reveals that evaluative structures are remarkably similar in these six societies, with a mean cross-culture correlation coefficient of 0.82 for social identities and 0.85 for behaviors. That means that people brought up in these Asian, European, and North American cultures largely agree about who is relatively good and who is relatively bad, and which actions are relatively right and which are relatively wrong. Japan is somewhat distinctive in the way it assigns esteem to people, and China is somewhat distinctive in its conceptions of morality. However, these Asian idiosyncrasies mainly are departures from the standards of European nations, and in any case none of the inter-cultural correlations involving evaluations go below 0.67. 

Notions of who is relatively powerful and who is relatively powerless also are similar across societies, though Germany departs a bit from the international congruence. However, consensus drops dramatically in judgments of which acts have potent impact, particularly for the two European cultures, which disagree even with each other. 

Ideas about who is relatively active or passive (i.e., who is likely or unlikely to initiate action) are moderately shared across the cultures, with an average inter-culture correlation of 0.61. Ideas about the spontaneity of actions also are somewhat shared, with an average inter-culture correlation of 0.59, excluding China. China is very distinctive with regard to assessing behavioral spontaneity, correlating essentially zero with other cultures. In part, this is because Chinese subjects rate interpersonal malevolence as non-spontaneous. For example, Chinese subjects rate 25 social behaviors more than 2.0 units lower in activity than do U.S.A. subjects, and all but one of these are bad actions. By contrast, only four behaviors, all good, are more than 2.0 units higher in activity. 

Overall, these results indicate that an international order circumscribes judgments of morality and the allocation of honor and stigma, power and oppression in interpersonal relations. Individual societies adjust this order in only limited ways. Greater diversity between cultures emerges with regard to the amount of agency allocated to different roles . However, the primary realm of cultural diversity is in definitions of the weightiness and spontaneity of social actions. 

On the whole, U.S.A. sentiments give better predictions of sentiments in other cultures than do sentiments of any other society. This perhaps reflects the hegemony of U.S.A. culture. Or it may be more accurate to speak of North America's hegemony, since correlations involving Canada are almost as high as for the U.S.A.

Cultural Stability

Cultures have another side (besides being homogeneous within a society and variable across societies): cultures continue for long periods of time - decades, or even centuries. Cultural change occurs slowly even in a modern society ridden with fashions, mass media, and social movements. At least so says theory.

We can examine this issue using two surveys of sentiments amongst University of North Carolina (UNC) students, one in the spring, 1975, the other in the fall, 1977.   Average female EPA ratings are available from both studies.

The second study was not an exact replication of the first, but there were substantial similarities. Though subjects in 1977 were different than those in 1975, subjects were selected from the UNC population of students in the same way both times. The two studies dealt with somewhat different sets of concepts, but 612 identities and 529 behaviors were common to both studies. EPA scales were defined with somewhat different adjectives in 1975 and 1977, but the 1977 adjectives were mostly a subset of the 1975 adjectives, as can be seen in the following table.

One way of assessing cultural change is to determine the extent to which later sentiments are predictable from earlier sentiments. Correlations measuring this kind of predictability are shown in the next table.

The predictability of sentiments over the two and one half year period is reasonably high in the case of identity EPAs and evaluations of behaviors. However, earlier sentiments are mediocre in predicting later potency and activity ratings of behaviors. Indeed, Canadian potency and activity ratings of behaviors ten years later predict the U.S.A. 1977 ratings better than the U.S.A. ratings from 1975 (see the cross-cultural correlation table above).

Rating scales were defined with different adjectives in the two studies, and the different scales may not be completely equivalent. In fact, further analysis reveals that ratings on the 1975 potency scale correlate with evaluation ratings, and that 1975 activity ratings correlate with potency ratings. These contaminations were the reason that the scales were changed between the two studies: the 1977 scales were designed to provide more independent measurements.

The non-independence of the 1975 ratings can be controlled statistically by regressing each 1977 variable on all three EPA variables in 1975. The second line of correlations in the table above are based on this adjustment, and these correlations reveal greater stability in potency and activity than at first seemed evident. Still, though, there is change in the potency and activity sentiments associated with behaviors. In fact, 1975 and 1977 U.S.A. behavior sentiments have about the same dissimilarity as do U.S.A. and Canadian behavior sentiments.

Could it be that the 1975-77 period was associated with exceptional change in U.S.A. culture? Here are events of national importance that were happening around that time (from The World Almanac and Book of Facts):

These events should have some relation to changes in behavior sentiments during this period if exceptional cultural change was occurring. The following table shows which behaviors stayed the same and which went up or down on potency and activity.

Over the 2� year period, a number of aggressive forms of deviance became more powerful, and a number of surreptitious forms of deviance became more active. Meanwhile, socially integrative acts seemed weaker, and intimacy behaviors seemed less alive. Maybe these shifts in sentiments did result from wrenching historical events of the time, though that is nothing more than a speculation to examine in the future research.

In any case, there is substantial stability in cultural sentiments about identities and a moderate amount of stability in sentiments about behaviors. Sentiment stabilities might be higher than this conclusion suggests if it is true that the one available stability study occurred at a time of transformation in U.S. society,

Instability or Unreliability?

Instability is one possible reason for imperfect correlations between measures of the same things at two different times. Measurement errors are another possible reason - correlations are pulled toward zero if measurements are imprecise. However, these two factors - instability and measurement unreliability - can be assessed separately if we have measurements at three times rather than two. And we do have one study in which EPA measurements were made at three times.

Undergraduates at a midwest university rated 50 identities related to authority, using computer-based continuous rating scales for assessing evaluation, potency, and activity.

The identities were: academic, advisor, airline pilot, attorney, authority, banker, barkeeper, bill collector, bodyguard, bouncer, champion, church deacon, confident, conformist, construction foreman, cop, counselor, dentist, detective, disciplinarian, employer, farm laborer, fireman, God, grandmother, headwaiter, hero, heroine, instructor, judge, juror, landlady, lawyer, manager, minister, mother, parent, policeman, probation officer, professor, prosecuting attorney, schoolteacher, sheriff, slavedriver, specialist, superior, supervisor, surgeon, teacher, tutor.

The first set of ratings by 44 females and 20 males was done in a spring semester, a second set of ratings by 38 females and 22 males was done eight months later in the fall semester, and a third set of ratings from 56 females and 15 males was obtained four months after that in the subsequent spring semester. Correlations among mean EPA ratings at different times were as follows.

If there are no measurement errors at all, then test-retest correlations are stability coefficients, and the correlations from time 1 to time 3 should be the product of the correlations during the two intervening times, reflecting the idea that more change occurs as more time passes. The diagrams above reveal that this is not the case - multiplying the correlations for the short periods gives a number far smaller than the correlation for the long period. So the test-retest correlations don't act much like stability coefficients.

Now suppose that the stabilities are close to 1.00, but there are measurement imprecisions causing test-retest correlations to be less than 1.00. In this case, it doesn't matter how much time passes between measurements: reduction in correlation due to measurement imprecision is the same whether the measurements are four months apart or a year apart, as long as no actual change occurs. This is the pattern we actually see: the correlations from time 1 to time 3 are of the same general magnitude as the correlations for the shorter periods.

The conclusion to be reached from this analysis is that the test-retest correlations are less than 1.00 mainly because of measurement errors. Correlations among Evaluation measures are not much less than 1.00, indicating that Evaluations are measured pretty precisely. However, correlations among Potency and Activity measures are lower, indicating these two dimensions are measured less precisely with a single rating scale.

The fact that male correlations are less than female correlations supports the conclusion that test-retest correlations are less than 1.00 because of measurement errors. Ratings were obtained from fewer males, and therefore male means have larger errors than do female means.

One problem with this analysis is that the identities were selected to study the meanings of authority, and, consequently, the identities are mostly good, potent, and quiet. Restrictions of range like this tend to reduce correlations. Another problem is that the test-retest correlations are computed over just 50 identities so we can't make too much of small variations in the correlation coefficients. A final problem is that we have no data at all for behaviors, settings, or modifiers.

Nevertheless this analysis does indicate imply that test-retest correlations are imperfect mainly because of measurement errors, and that fundamental sentiments are quite stable over periods of a year.

Conclusions About Stability

This got complicated!

The first analysis above suggests that a notable change in sentiments occurred in American culture between 1975 and 1977. Yet the second analysis suggests that over-time correlations less than 1.00 occur mainly because of measurement imprecision, so maybe there wasn't a big shift in culture during the 1970s. On the other hand, the second study didn't look at over-time correlations for sentiments about behaviors, so maybe the meanings of behaviors did change in the 1970s as the first study suggested.

Well, that's the way the knowledge business works - we need more research to sort this out.

At least we know now that evaluations are pretty stable over a period of a year or two, and that cultural change might impact on sentiments about behaviors more than on sentiments about identities.