Kaz's SAS, HLM, and Rasch Model Statistics Tutorials: 3 approaches to learning
 Home Large National Data Sets Kaz Library SAS manuals What is "error"? Rasch Model HLM SAS PROC NLMIXED and GLIMMIX Factor Analysis Reading output via SAS Excel functions for Statistical Analysis Essays on learning statistics Statistics in Japanese My profile My SAS questions and SAS's responses My work tool box

My lectures on Statistics

ExperienSTAT 1.1

Learn Basic Statistics by "doing" them

I prepared these for TAing Social Statistics at the University of Chicago in the last century.  I wrote most of them in order to teach with a textbook Statistical Methods for the Social Science (second edition) by Alan Agresti and Barbara Finlay (1986 Dellen Publishing Company).  I used this for TA sessions on Friday afternoon in a computer lab.  Students by then were exposed to concepts in class.  I wanted them to experience the concepts by doing them.

I felt that my students were pretty much engaged in this activity.  I think I did not have to speak too much.  I just explain basic things about the excel sheet and let them go loose.  I think it engaged students because my students were anyways smart and they liked figuring out things on their own.

Unfortunatley these materials are not self-explanatory.  Please let me know if you can collaborate with me to make them more user-friendly.

Learn OLS by using SAS IML (Interactive Matrix Language)
Excel is fine, but there is a limitation.  We want to experience matrices too.  I think doing so will help us recognize how parts are related in the whole of a statsitical model, like OLS.  ALso there is a virtue of learning PROC IML, which is extremely useful for doing social network analysis, for the calculation of network constructs, such as density and centrality.
The coefficients for an OLS equation are obtained by only one line, "beta=(inv(t(x)*x))*(t(x)*y);"  And I think there are about five or more lines to get standard errors for the coefficients.  This section for standard errors ends with covariance_of_betas=inv(t(x)*x)#mean_squared_error;"  You can know what "covariance-variance matrix" feels like and looks like.

Before this exercise I had no clue why my professor called it "variance-covariance matrix."  Well, standard errors and covariance of those errors come in a matrix in this way, which is why it is called that way.

Finally, the bonus of trying this is that you will learn how each section is related to each other.  Before this exercise I did not have a strong sense of "sections" in the series of equations that a professor wrote on the blackboard.

For example, you know the line for coefficient estimates MUST come first before anything else.  Things about Variance should come before the estimation of standard errors, for example.  You begin to see the connection among statistical concepts, such as standard error, t-score, p-value, R-square, etc.

Use SAS PROC NLMIXED to learn about statistics in general
PROC NLMIXED is the most general statistical procedure in SAS.  "General" means that it can do almost anything, including getting a simple mean to doing a complicated thing like hierarchical linear model or Rasch model.  Although these different statistical models have different names, they are of the same gigantic statistics family.  By forcing oneself to use PROC NLMIXED and try different models, one will understand how one model is related to another.  You will enjoy the process of finding how, for example, a simple average, is a special case of a linear model.  PROC NLMIXED also forces you to use only one way of writing equasions, so you won't get confused with different styles of writing equasions that people do when they teach.

 Statistics with hands-on Copyright 2005 KU For information inquiry (AT) estat.us