mean +/- 2 std

[Doris Rivera]

Dear Statalist, I am dealing with a variable that it is in between 0-1 (taking any possible value within these bounds). This variable serves as a moderator (interaction) in a regression. I was asked to report the marginal effects for the mean +/- 2 std (assuming that 95% of observations will be in between). However, as you can see next, the mean – 2 std would be a negative value (which is outside the range of the variable). Does this have to do with the fact that the variable does not follow a normal distribution and that it is not symmetric? Or is it something that I am missing here?

I know it is not properly a Stata question, so, I apologize in advance for the inconvenience.

Thanks for your help.

-------------------------------------------------------------
      Percentiles      Smallest
 1%            0              0
 5%            0              0
10%            0              0       Obs              47,493
25%            0              0       Sum of Wgt.      47,493

50%     .0094595                      Mean           .0505029
                        Largest       Std. Dev.      .1117768
75%         .044              1
90%       .13755              1       Variance        .012494
95%         .261              1       Skewness       4.052373
99%     .5909091              1       Kurtosis       22.93133

[Andrew Musau]

mean +/- 2 std (assuming that 95% of observations will be in between)

This rule of thumb applies to a variable that is approximately normally distributed. For such a variable, skewness=0 and kurtosis=3. This is not the case for your variable looking at the descriptive statistics. There is a skewness-kurtosis test of normality

sktest var

or you can graph to see this:

hist var, normal
qnorm var

where you replace "var" with the name of your variable.

[Nick Cox]

A normality test or check is unlikely to help much. For example somewhere between 25% and 50% of your data are exact zeros. But values of 1 occur too, so a grid of points on [0, 1] might be advisable.

[Doris Rivera]

Thanks both for the advices. At least I was not so wrong. Yep, it would be better to report the marginal effects for several points.