Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | Jessica Chan <jchan87012@gmail.com> |
To | statalist@hsphsun2.harvard.edu |
Subject | st: Scientific notation in regression estimates (e.g., 2.5e-16) |
Date | Thu, 11 Jul 2013 14:23:50 -0700 |
Dear Statalisters, My quantile regression output is displaying very small coefficients (basically 0), and displaying these in scientific notation format. 1. Does this reflect an error in the modelling or estimation procedures? Do these represent perfectly valid estimates, or should I be wary of using these estimates? Note, I'm using quantile regression, so I'm thinking this has something to do with estimating the medians rather than the averages (e.g., due to precision, the regression results are returning, for example, 0.0000000015 rather than 0). Here is a fictitious example: sysuse nlsw88 qreg age occupation qreg age race qreg age occupation race *occupation is entered as a categorical variable for illustration purposes *in the first model, note that age and occupation are unrelated, as expected, and that the coefficient is 0 (i.e., no difference in median age between occupations) *in the third model, note that the coefficient for occupation is very small and in scientific notation (but given the results of the first model, I think this is fair to assume that it represents 0). 2. Separately, how should I interpret regression output that has missing t or z statistics, and "point estimate" confidence intervals? In the above example, I note that the standard error is present, so I'm thinking this is just a limitation of the output display and that a very large value for the t or z statistic (i.e., generated by dividing the beta estimate by the very small SE) is displayed as "." instead of the large value (e.g., 5e10). Thanks for your help, J * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/faqs/resources/statalist-faq/ * http://www.ats.ucla.edu/stat/stata/