Power analysis probit model

[Aleksander Gwizdak]

Hello,

I wanted to ask how to assess the statistical power of a probit model. In my dataset, the dependent variable is equal to 0 in 726 observations and equal to 1 in only 30 observations. Is there any way to verify if this smaple is big enough? I am rather new to STATA and would appreciate any help. Thanks in advance.

Olek

[Carlo Lazzaro]

Aleksander:
a similar topic is covered at: http://www.ats.ucla.edu/stat/stata/dae/logit_power.htm

[Aleksander Gwizdak]

Thank you for the link. In my dataset however the main independent variable is binary and I cannot use powerlog. Is there any way to deal with this issue?

[Carlo Lazzaro]

Aleksander:
for the following toy example, that mimicks the one reported in the link I pointed you out to in my previous reply, let's assume that:
-your p1 is: (500/1000)=0.50 and its standard deviation is: (1000*0.50*(1-0.50))^.5=15.81;
-your p2 is: [(500+15.81)/1000]=0.516:

powerlog, p1(.5) p2(.516) alpha(.05) help

As an aside, what sounds critical to me considering the details you provided in your first post, is that you have a very low probability [30/(726+30)]=0.04 of obtaining 1 (and, conversely, a very high probability of obtaining 0). Does this feature of your data give you perfect prediction issues?

[Aleksander Gwizdak]

Thank you very much, now I understand it. It actually does not give perfect predition. The regression that I use is only a sideline regression (in my thesis) and I want to test the statistical power as the results are not as I expected them to be. I want to present some quantitative evidence for disregarding the results.

[Carlo Lazzaro]

Aleksander:
thanks for closing this thread positively.
All the best for what lies ahead your data analysis!

[Gizem Turna]

Hello,

I have an ordered probit model summarized as follows:

Independent variable number of obs.
Y=0 117
Y=1 292
Y=2 102
Also, the regression has 13 explanatory variables in total, and the number of observations in the regression is 442.
I want to find the power of this model. How can I calculate it with Stata 15?

Thanks in advance,
G.