Interpretation

[Guest]

I have the below model, how should I interpret B6: There is a decrease in children by 0.21(It does not make sense???) if an African- American woman live in urban area than rural area, other thing constant and it is significant(p-value:0.000).

. regress children age agesq spirit protest catholic urban educ elec educ_elec, robust cformat(%9.3f)

Linear regression Number of obs = 4358
F( 9, 4348) = 648.14
Prob > F = 0.0000
R-squared = 0.5742
Root MSE = 1.4514
------------------------------------------------------------------------------
| Robust
children | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
age | 0.343 0.019 17.90 0.000 0.306 0.381
agesq | -0.003 0.000 -7.91 0.000 -0.003 -0.002
spirit | 0.135 0.056 2.40 0.017 0.025 0.246
protest | 0.071 0.066 1.07 0.283 -0.059 0.200
catholic | 0.116 0.079 1.47 0.143 -0.039 0.271
urban | -0.210 0.046 -4.56 0.000 -0.300 -0.120
educ | -0.072 0.007 -10.00 0.000 -0.086 -0.058
electric | -0.129 0.183 -0.70 0.482 -0.489 0.230
educ_elec | -0.022 0.018 -1.18 0.237 -0.057 0.014
_cons | -4.360 0.250 -17.45 0.000 -4.850 -3.870
------------------------------------------------------------------------------

[Carlo Lazzaro]

Guest:
it makes sense just like stating that one-year increase in education reduces the number of children (-0.072 and significantly so) when adjusted for the other predictors.
The main issue here is that you decided to apply a linear regression model to analyze a regeressand that is not continuous, but can take on integers only (1 child, 2 children and so on).
Hence, the advice is to consider -poisson- (and, if overdispersed, -nbreg-) for your analysis.
As an aside, please note that creating categorical variables/interactions (linear and squared terms for -age- in your model) by hand is far less rewarding than applying -fvvarlist- notation, that has also very good relationships with two wonderful Stata commands, such as -margins- and -marginsplot-.

[Weiwen Ng]

Per Carlo's comment about the factor variable syntax, he means that you could have typed something like this command instead (taking guesses as to which are dummies you created and which variables are continuous vs categorical):

regress children c.age## c.age i.spirit i.religion i.urban c.educ##i.elec, robust cformat(%9.3f)

This is much less effort than creating your own dummies and interaction terms, it will reduce any errors, and it is guaranteed to work properly with the margins command.

Your interpretation is roughly correct. A better statement might be that on average, urban women have 0.210 fewer children than rural women, controlling for everything else. If this still does not make sense, can you clarify why?

[Carlo Lazzaro]

Guest:
while I do share Weiven's helpful advice, due to the nature of your regressand I still believe that you should switch from -regress- to -poisson- (and -nbreg- after a probably overdispersed -poissson-).