A reflection on expected experimental outcomes – Latent Semantic Analysis of interviews

When I first thought of the possibility of using Latent Semantic Analysis (LSA) to analyse my parent interviews exploring school choice my expectation was that the odds of success were slim.  I was familiar with the use of LSA in helping decipher the contextual meaning of words in ancient texts and indicate the likelihood a text was written by a particular person based on their known works.  LSA is one of the methods used to try and identify (speculate) who the ‘real’ Shakespeare was by comparing the works of Shakespeare with the writings of other contemporaries.  But could LSA be applied to interviews investigating economic decision making behaviour and provide meaningful insights? Would it be possible to identify key concepts influencing the decision making process based the common use of key words? Could LSA identify ‘latent authors’ representing distinct, heterogeneous, types of decision making within a society? Noting that the conventional economic wisdom is that society is comprised of a homogenous set of individuals (one type) applying the same decision processes subject to variability environmental conditions (such as wealth) and uncertainty.  To draw on quantum physics – there are no ‘flavours’ or handedness in standard economic theory.

So – I knew the maths and theory of LSA was solid but quite uncertain as to whether it could be applied to interviews into economic decision making.  I rated my chances at best 1 in 10 (10%) of finding heterogeneity, ‘flavours’, in the interview data collected.  My low expectations of success are largely due to a belief that individual diversity, the existential uniqueness of each individual’s circumstances and personal history, would overwhelm any attempt to categorize individual interviews into groups.  The decision to invest a fair bit of time learning how to program and implement LSA in Python was a bit of long shot gamble where the low odds were offset by the reward of potentially exciting results.  My highest realistic hopes though were to find a level of heterogeneity separating parents of Asian and European descent by differences in culture – particularly given that Asian cultures tend to ascribe a much higher importance to education than their European counterparts.  Finding other ‘flavours’ or ‘handedness’ in economic decision making processes types would be truly exciting.

The valuable insights that linguistic analysis of decision making processes can give can be seen in the research done by Jonathan Haidt.  He has found that heterogeneity in moral and ethical reasoning is reflected in different word use between religious and political groups within and across cultures.  Two of his most cited papers are given below.


Haidt, J. (2001). The emotional dog and its rational tail: a social intuitionist approach to moral judgment. Psychological review108(4), 814.

Greene, J., & Haidt, J. (2002). How (and where) does moral judgment work?.Trends in cognitive sciences6(12), 517-523.