First thoughts on Induction and Statistics

A friend recently asked me to comment on Whewell’s theory of induction.  Below is my response to him with some slight edits.

In some ways, Whewell’s theory is a bit alien to me because I’ve been introduced to modern research methodologies that use different terminology.  If I studied his writing in more depth, I could probably find similarities between his work in induction and my training in statistical analysis – including things like establishing reliability and validity through statistical tests and their relationship to Whewell’s notion of “colligation”.  These statistical tests check to see if relationships between various measured constructs are more than just pure chance.  Part of my training was to be clear that finding such relationships DOES NOT demonstrate causality, but perhaps influence or impact one another.

In the business world, it’s often difficult to determine things with absolute certainty.  For example, a businessman may want to know the major factors that influence purchasing products online (something that I’ve researched).  This obviously varies significantly from person to person (and may change over time), but the businessman wants to know what areas would help them to focus their efforts right now.  So they want to know which of say 10 things are the most important.  They can’t survey everyone, so they take a sample of the total to get some idea.  Using statistical techniques, they can determine if the sample is big enough to be meaningful and they use different techniques to establish if the sample results are reliable and valid.  It’s important that the observations provide some new knowledge that is justifiable and useful. This is significantly different than what happens in the hard sciences where hypotheses can be experimentally tested and verified or rejected.

To relate it to some of my recent work, I did a project on job seeker response to privacy endangering request for social media login information (Drake, Hall, Becton, & Posey, 2016). In it, we created hypotheses such as: “Job applicants who judge a request to share social media login information as immoral are more likely to engage in information privacy protective responses” We checked six different privacy protective responses to discover which was preferred.  In our discussion, we state that we found evidence that the strength of the moral judgment impacts five of the six privacy protective responses. I’m not entirely sure how to phrase these findings in terms of Whewell’s theory.

I tried reading a modern philosopher, Jonathan Cohen, and his theory of induction, but couldn’t get very far.  He tackles induction with statistical analysis, which I hoped would enlighten me to the difference between modern techniques and more classical theories of Bacon, Whewell, and others. But it seems I need to read more about them first to appreciate Cohen’s work.

One criticism I have of Whewell after reading some summaries of his theories, he seems influenced by platonic forms in his notion that ideas, especially fundamental ideas, exist prior to experience. But it seems to be a weak form of rationalism, as most of this emphasis of induction is using observation to “unfold” the concepts and ideas.