Sunday, October 28, 2012

Monday, October 15, 2012

Wednesday, October 10, 2012

The Importance of Questioning Correlations

On the topic of correlations....this story came out a couple of weeks ago, about a recent study linking equal division of labour within couples to higher divorce rates. The most interesting thing about this is the way the study was reported in the media - with headlines like:

Significantly, rather than jump on the premature-conclusion-bandwagon, many other journalists questioned the study's conclusions, pointed out that correlation is not causation AND considered the very compelling possibility of a spurious relationship (in which "traditional values" becomes an overlooked, potentially confounding, variable). For example, as Wendy Leung wrote in the Globe & Mail:
According to the Telegraph, a new study out of Norway has found that divorce rates are as much as 50 per cent higher among couples that share the load equally, compared with households where women do the majority of the chores. But wait a minute. Aren’t households in which women do all the cooking and cleaning simply more traditional, and less inclined to see divorce as an option? The researchers acknowledge this could be one explanation.
FYI, a "confounding variable" is something that obscures the relationship between an independent and a dependent variable - in this case the relationship between "shared housework" and "divorce rates".  The argument Leung and others make is that "traditional values" may also be impacting upon both divorce rates AND housework distribution in significant ways that are overlooked in simplistic comparisons of housework and divorce. i.e. couples with more traditional values are less likely to get divorced AND less likely to share the housework equally. If a number of such couples were included in the study, how does this problematize the conclusion (or more accurately, the newspaper headlines) that sharing housework "leads" to divorce?

The sloppiness of the study (& its conclusions) itself has been noted by several journalists, but this article by Jen Doll in The Atlantic delves into things a bit more in depth, reviews some of the relevant academic literature (much of which deeply contradicts the idea that housework causes divorce), interviews with experts, etc. A really key point that Doll makes is that other than describing the country of origin, few if any of these news reports pinpoint the resesearchers' affiliation(s). This is a pretty crucial omission - is this a university study? A study conducted by a particular organization, charity, church, think tank, political party??? An informal survey some guy did of his friends and family???? Fascinating to see how studies like these end up as news headlines, while others (most) are ignored completely. For INF1240 it's also interesting to see how a deeper knowledge of research methods (incl. research design and data analysis) can be useful in questioning the research AND how it is portrayed in public/news discourses.

A Very Brief Intro to Correlations

Via Sociological Images, a great graph (any clear, easy to read graph is pretty great) and interesting analysis of the *positive correlation between income and SAT scores, from data published by The College Board. There's a pretty strong relationship implied here - one that raises questions about the ongoing reproduction of class inequality and the hidden bias of standardized tests (as discussed briefly in relation to IQ tests - see Stephen Jay Gould).

©2010 Sociological Images

* Positive Correlation: Defined by Timothy C. Urdan as: "A characteristic of a correlation; when the scores on the two correlated variables move in the same direction, on average. As the scores on one variable rise, scores on the other variable rise, and vice versa." (For more, see Urdan, T.C. (2005) Statistics in Plain English (2nd edition). Lawrence Erlbaum Associates, Inc.)

Note: Positive/negative correlations are found through "Correlational Analysis," which measures the strength of an association between two variables. Values range from +1.00 to –1.00. 

Rule of Thumb: Correlation should NEVER be confused with causation = they are very different things, and involve a very different set of calculations and often different research designs (methods, analysis, control groups, etc.). Causation causes correlation, but it is not necessarily the other way around. It is much easier to establish correlation than causation. And it is also very easy to confuse or inflate the significance of correlation - as seen in the media effects debates discussed in next week's Kline reading.