Recent content by noetsi

It may attenuate the slope (at least if its a dummy predictor although I think 90 percent at one level is where it gets dangerous). According to various authors there are minimum requirements for the least common level of the dv although various authors disagree on them and they are rules of...

There are various rules of thumb and various authors disagree about them. Commonly they focus on the least common level of the dependent variable. However, some say if you have a dummy predictor with (I think) 90 percent or more of the values at one level it will attenuate the slope. If I can...

A basic problem with stepwise is coincidences or very small differences can make a huge difference in your results. And if what you leave out of the model this way is nearly as important as one of the variables you include and correlated with it you will bias the slope coefficients.

It commonly depends on if your clusters or groups actually matter I believe. If they explain little variance I suspect that the linear and multilevel results will be very similar even if formally the linear regression violates the assumption of the method in the case you describe. Formally count...

A statistical control does predict the dependent variable. That is why its in the model. If your control explains most of the variance not your other variables than it does - and that is worth noting. Things beyond policy makers control are commonly critical in the economic literature.

"In this problem, there is no unknown; you know the real values, if they aren’t the same, then they’re different." Amen. This should be emphasized in any statistics course when they get to p values. Its amazing how few get it.

First forget stepwise. Its a bad idea. Second, if you have a theory put in variables based on that theory. If you don't, and assuming that you have enough data for decent power run all the variables. I would toss the ones that are not significant and run the model again without these. Do the...

I think you could do a test of proportions if you want a statistical test. But as I and hlsmith noted it is not really valid to use a statistical test for what you want. Statistical test make sense when you don't know the true effect size. Here you do know it.

Archidamus I think the state average is known (it's 68 percent). The question asked is if any compliance is significantly different that this. While you could do tests like test of proportion to answer this, my argument is that this is really a substantive question that (as is often done, but...

I think a good starting question is why do you need to do that. Test of significance get at whether the effect size you find is likely to exist in a population when you have a sample - that is it is the true effect size. It does not tell you that the effect size is substantively different than...

Its very difficult to get at "explained variance" in logistic regression the Pseudo R squared don't show you that. How are you determining explained variance in your model? If you have a model that is significant, but no variables that are then Multicolinearity is a likely issue. I think if...

Your results make no sense to me, how you code something has nothing to do with its significance. I can only assume an error in the coding or data or possibly power issues.

That is probably true, but I would bet 1 is still in the CI.

The tables are lousy. Better ones generate the slopes and p values (these simply tell you they are sig at a certain level and then only in the notes).

"By default SAS assigned binary text variables with values -1 and 1." From memory that is effect coding. Personally I would use reference coding where the values are normally 0 or 1. This is the most common way analysis is done.