12
Stata Technical Bulletin
STB-33
Example
The following uses the example given in Stephens (1986):
. input u
u
1. .004
2. .304
3. .612
4. .748
Б. .771
6. .806
7. .850
8. .885
9. .906
10. .977
11. end
. nbgof u
Neyman-Barton
Variable Smooth GOF Test р-value U-bar S-squared
---------+------------------------------------------------------
u I 6.437 0.0400 2.041 1.507
References
Berry, K. J. and P. W. Mielke, Jr. 1994. A test of significance for the index of ordinal variation. Perceptual and Motor Skills 79: 1291-1295.
Stephens, M. A. 1986. Tests for the uniform distribution. In Goodness-of-Fit Techniques, eds. R. B. D’Agostino and M. A. Stephens. New York:
Marcel Dekker.
sg60 Enhancements for the display of estimation results
Jeroen Weesie, Utrecht University, Netherlands, [email protected]
The regular Stata output of estimation commands comprises parameter estimates, standard errors, z or t statistics, ^-values,
and confidence intervals. Clearly, there is a lot of redundancy in this information. For instance, z and s statistics are simply the
ratios of the estimates and their standard errors. (Note that this is not fully correct: In exponentiated form, z and s statistics are
not transformed by Stata.) Hypotheses testing is possible either via confidence intervals or via ^-values.
In practice, many researchers only consider a few of these numbers, in particular the parameter estimates and the associated
р-values. Thus, precious “display space” seems to be wasted. Indeed, at the same time, Stata’s regular output does not contain
pieces of information that I find quite useful.
First, Stata’s variable names, as are all of its identifiers, are restricted to length 8. In many cases, this is hardly sufficient to
produce meaningful names. For instance, in many surveys, variables are named V013aj, etc. Additionally, the names of variables
produced and named automatically by programs such as xi are hardly more understandable than assembler mnemonics. Variable
labels, even those produced automatically by xi, are usually easy to understand. These labels could simply be included in the
output.
Second, to interpret the “size of effects” it is useful to see the location and scale of variables along with the parameter
estimates. This practice is followed by many statistical programs including SPSS, BMDP, and LIMDEP.
This insert describes a program diest, that can be used after any Stata estimation command such as regress, logistic,
or heckman. Note that diest also should work properly with multiple-equations models such as mvreg. diest redisplays the
table with information about the parameter estimates (not the parts above and below the table, such as the number of observations,
the log-likelihood, etc.). This table always includes the variable names, the variable labels of the dependent and independent
variables, and the parameter estimates.
Via options, the user can select additional information, such as the standard deviations, confidence intervals, or summary
statistics of the independent variables. In addition, to facilitate the inclusion of Stata output in reports that describe statistical
analyses, we provide a series of options that specify display formats.