Coming in Stata 13, good news for users who need to export results without hacking too much into the software (rejoice, students!). This reminds me of Tamás Bartus’ publish command, which has been in development for several years now.
This also confirms that Stata aims at users who program a bit before going back to their favourite WYSIWYG software, while R aims at users who program a lot from their favourite code editor. There is demand for both.
Stata 13 will be out in a few days and also contains good news for structural equation modelers — one more indication of the software’s academic target audience.
Carcere Rodeo: Thoughts on binary Ys: Mostly a response to Bellmare, mostly for @jay_ulfelder
prisonrodeo:
So this is primarily a response to this; if you haven’t read it, do so first.
On his first three points:
1. Robust standard errors don’t do what we tend to think they do. See here: http://j.mp/InK5jU . Moreover, if we know what the right form for the variance is, why not use it?
2. Forecasting probabilities / estimating predicted probabilities — and the changes we expect to see in them as the result of D — are in fact exactly what the readers of that blog presumably want. The parameter on D (\gamma) is of direct interest only in the linear case; otherwise, we typically want something that is interpretable on the “scale” of Y. We can get that — easily — without resorting to a model that does violence to the functional form of the relationship (see point 3, below). More specifically: if I told you that the E(Y|D=0) was 0.7, and E(Y|D=1) was 1.1, and you knew Y was binary, would that not be a problem?
Regarding the final remarks of the response, can I join the “always logit, never probit” team without providing a full-fledged argument?
One-line summary statistics table: a wrapper for estpost
Some time ago, I released a wrapper for the catplot command to produce bar plots of categorical data in one short command. I did not find the time to improve it, so I am just adding a little demo to it and calling it quits for now. Here’s the code.
I just wrote another little wrapper to produce simple summary statistics tables in plain text format with tab-separated values. I ask students to produce such tables but was not happy with existing commands like tabout and tabstatout.
It’s a bit of a paradox that most statistical software do not produce these “Table 1” outputs in a straightforward manner. My need was for a result that would be easy to copy-paste and then format in an editor like Google Documents or Microsoft Office.
With that aim in mind, the wrapper uses the estpost and esttab commands from the estout package to export a mean/SD/min/max summary for continuous variables and percentages for categorical ones, using complete observations only and fixed precision at zero decimal, except for the mean and SD that get one.
A previous version of the wrapper did a few more things such as accepting a by(varname) option, but at the price of much clunkier code. This is what the output of the newer version looks like in the Stata Results window:
The stab_demo.txt file returned by the wrapper contains the same information, minus the blank lines. The demo code follows, where m() stands for mean and p() for prop (proportions).
webuse nhanes2, clear
la var sex "Gender"
la var race "Race"
noi stab using stab_demo.txt, m(age height weight bmi) p(sex race) replace
Reminds me of the map of Istanbul that was on TV last night.
