Under Construction

I haven't really begun to build this page.
I started my career a statistician, so its a subject I'd like to go back to.
With all the current emphasis on data mining and building profiles (targeted advertising, security risks, product safety, drug benefits, credit ratings, etc.) based on statistics, it is all the more relavent.

Multicollinearity:
One area in statistics that constantly bothers me is researchers who draw cause/effect conclusions when two variables are correlated.
While in grad school at Cal, I took a linear regression class from a world renown statistician. This was his pet peeve also. I saw professors and grad students making this mistake all the time.

The common example used to illustrate this is the correlation between ice cream consumption and drowning deaths, which are very correlated in time series data.

It is a multicollinearity problem.
Both variables are correlated to a third variable, temperature, which was not modeled.
See: Correlation, causation and forecasting | OTexts

My personal gripe is people telling me, a widower for many years, that I would live longer if I got married again. Statistics show this.
Someday I'm going to get the data on health, which is omitted in these studies, and see what it shows.
I think unhealthy men (overweight, smoking, addictions, ...) are less likely to find a partner and also die younger.
Some day I'm going to write a page on the pros and cons of marriage after the kids are gone.

Another one I'd like to research is the apparent correlation between evangelical Christians and people who like Donald Trump. That's a real sticky wicket.
See Christian Right and Republicans

Links:
Amazon.com: How to Lie with Statistics , Darrell Huff
Correlation, causation and forecasting | OTexts
R Statistical Programming Language

last updated 3 Oct 2016