[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
Re: st: Stata v. SAS OLS
If you have Stat Transfer available, you could run the following experiment:
-- transfer your SAS data set to Stata & run the regression in Stata
-- transfer your Ststa data set to SAS & run the regression in SAS
I would think this would reveal something.
>I am having some trouble trying to match OLS regression output done in
>SAS to a OLS regression done in STATA. Basically, I was given a SAS
>program, and when run on SAS, it produces the exact results we are
>looking for. However, I'm trying to "convert" it into a STATA program.
>The program itself appends and merges a number of datasets, creates some
>variables from the final merged dataset using some simple arithmetic
>commands, drops certain observations, and then runs a OLS regression on
>certain variables. It is really not all that complex of a program, but
>I cannot for the life of me get my STATA regression results to match the
>SAS results. This is made even more bizarre by how similar the final
>datasets appear to be. All of the summary statistics (i.e. mean, std.
>Dev., max, min, number of observations) are identical in both programs.
>Moreover, I generated an "id" variable that simply numbers the
>observations in the first (and largest dataset) in both programs, and
>after all the datasets are merged and transformed, both datasets contain
>the exact same observations from that first dataset (e.g. the id
>variables from both programs are identical). When I run the regressions
>in STATA, the coefficients for the variables all come very, very close,
>if not exactly the same as the coefficients in SAS. The only exception
>to this is the constant, which bizarrely is nearly 100 times larger than
>the constant in SAS. And I cannot figure out why! Does anyone know
>what would cause supposedly identical datasets to produce such different
>results in the constants? I have a pocket theory that it is because the
>programs treat blanks and zeros differently, but I have been unable to
>prove it, and the problem could very well be something else. If anyone
>can offer any help, I would greatly appreciate it!
Eric G. Wruck
2535 Sherwood Road
Columbus, OH 43209
* For searches and help try: