TO: Students in CJ 405
FROM: R. B. Taylor
DATE: 10/15/01
RE: Comments on homeworks completed for 10/8

Here are a few comments on what seemed to be some of the most typical ways to go awry in completing the homeworks you handed in on 10/8.

EXPLAINING VS. DESCRIBING. When I ask you to interpret the relationship you have just described using b and beta and a, I am asking you to not just describe (e.g., as imprisonment rates go up, so too do violence rates several years later). Of course describing is the first step. But when I am asking you to interpret I am asking you to give me your thoughts about what dynamics might be involved in, or might undergird this relationship. Talk to me about dynamic processes.

THINKING ECOLOGICALLY. It is important when you are working with the ecological data set that your explanation be framed at the ecological level. Do not talk to me about individuals who do this then do that. Rather talk to me about states that have this kind of feature {describe} are likely to have that kind of feature {describe} and here are the reasons why. Think about ecological attributes. When you are thinking like this you may wish to make inferences from observed variables. For example, you might decide that the rate of divorce is really a proxy variable for the broader concept of anomie, which of course you would then go on to describe.

Some folks committed the ecological fallacy (thinking that ecological variables describe individual level relationships) when they  talked about people going into prison and being divorced, for example.

PAY ATTENTION TO THE TIMING OF THE VARIABLES. In the case of the ALLBOOZ (1983) variable and the HSRATE (1994) variable, those students who graduated from high school in 1994 were about 7 when their or other people's parents or just plain adults were out buying alcohol in 1983. When describing the state level dynamics.

BETA turned out to be hard for a lot of folks. To review:
* both variables are z scored so now a one unit change on each = 1 standard deviation
* so if beta = .57 and you are looking at the effects of a one standard deviation increase in the predictor resulting in an increase of .57 standard deviations (57% of a standard deviation) on the outcome; cases that are that much higher on X (one sd) will be that much higher (.57 sd) on Y; at least that is what is PREDICTED, although of course the residual may not be zero