Discussions, Interviews & Lectures Detail
:: Description :: Alan Lomax, David Brown, and Norman Berkowitz discussing geographic factor analysis and expressive style: the validity of an approach using small samples
:: Project :: Cantometrics
:: Date Range :: 03-19-1976 to 03-19-1976
:: Particpants ::
Lomax, Alan
Brown, David
Berkowitz, Norman
:: Subjects ::
Geographic factor analysis of expressive styles
Small samples - validity of
George Murdock's social variables giving too much weight to the family
Dance style and its relation to technology
Stratified versus random samples
Prive and public song styles - resemblance between indicate that style supervenes function
:: Cultures ::
General
:: Holdings ::
:: Notes :: T3643: Propriety of allowing factor analysis to simply delineate highly inter-related clusters and then to drop the loadings. Tests of significance and subsequent cluster mapping. Clusters versus weighted linear combination variables. Uniques and clusters. Indicating strength of link with bandwidth graphic - a diagram that summarizes much. Alan Lomax: The dimension of complexity brings together all similar factors from all the systems. In a case where one dominant measure is emerging, it is insensitive to minor variations. Standardizing the range is similar to turning to two scores. Significance of a dual approach: first take a carefully ordered world sample of cultures that represent all the provinces of the world, stratified, in that each province has about the same number of cultures. We get six regions and each region has, say, 10 areas per region (e.g., the plains and the pueblo). Six regions produce 57 areas. From a second factor analysis of these we get zones of culture all geographically pure. A bonded map of those gives about the same neat picture of human evolution that we get from this other and independent analysis. This augurs well for the use of limited resources. Testing parsimony. Doing it two ways is a substitute for a significance test. A somewhat different sample gives the same result. We've gone one step further. We have done a separate factor analysis of geographic taxonomies. Two different systems of measurement for social, musical, dance, instruments, and speech produce maps that are alike - not the same but close family resemblance. An added feature: you can identify variables that have similar meanings. Ecological discoveries have been made independently from different centers. Does lack of agreement in geographic clusters reflect different proportion of some dimensions than they do other elements? About 25% of variables reflect complexity - stratification, whereas in song 87% reflect these dimensions. Perhaps we should have fewer domains. Complexity loading has to be very significant. Alan Lomax: The social variables in Murdock were arrived at ad hoc. They concentrated so heavily on the family that we had to throw out about a third of them. The variables were biasing the result. The problem is how to have a balanced system. Choreometrics, on the other hand, totally weighted every factor to technology. Of ten common factors, two thirds were tied to manipulation. Margaret Mead's comment was: "Of course, dance is very dull." When we did regional analysis by song we had to change the weight of complexity to make the neatest possible regional geographic taxonomy. Everything in Europe turned out to be the same because we had so many European measures. We reduced the value of ten of those measures. In other parts of the world we weighted complexity up because we had so many measures of their simplicity. Data reduction. R-factors and Q-factors. Our approach has been to reduce the enormity of the data. For example, instead of "embellishment," ornamentation in general. Norman Berkowitz: It's not really a reduction but an effect of a heap of factors on another heap. Most factor analysts posit that one can take single values. If you were to take one song, it represents the culture but ten songs would be better. If you mix together variables it should be more stable. Alan Lomax: By cutting the factor fields and comparing the differences we might get a better, simpler structure. T3656: Stratified vesus random samples. In some cultures more people walked in front of the tape recorder than others. Public songs versus private (sacred songs). Norman Berkowitz: Do cultures have sacred songs that they don't sing to outsiders that will be under-represented? Alan Lomax: I don't think that the songs differ in style. A basic finding of the study is that style supervenes function - with partial exceptions. Cantometric areas more reliable than Murdock Provinces. Gold analysis (analyzing frequency of occurrence of a sample that has been selected to have equal size in every province) versus simple analysis (finds mean occurrence, does analysis of these). How to check a smaller sample for accuracy: take 20 samples, split them in half and compare modal frequencies. Significance of first case versus subsequent cases. Checking through random resorts. (Alan Lomax: None of the people who object to the system would understand it.) Adjunct cultures. Weight the experiment against you. If you limit the randomness it would weight it against you. Prune the less useful samples. Set it up so that it can be used with, say, a week of training. A few can use all of it, but many could use some of it.

 

 

© 2001-2009 Association for Cultural Equity | Contact | Credits | Rights