Standardized Coefficients

[stephlp]

I'm currently writing my dissertation and I'm in the process of writing up my results. In a few studies, I have seen people report the results as 'a one standard deviation increase in the independent variable increases the dependent variable by x%'. I have also seen people report the results as 'a one standard deviation increase in the independent variable increases the dependent variable by x standard deviations'.

So for example, one of my regressions is

xi: reg GrowthrateGDPpc SecondaryaidGDPperYear Govt CapitalFormation lnin lninitial trade i.Year, beta

GrowthrateGDPpc	Coef.	Std. Err.	t	P>t	Beta
SecondaryaidGDPperYear	-6.81619	4.070058	-1.67	0.096	-0.12313
Govt	-0.0961052	0.0572342	-1.68	0.095	-0.11855
CapitalFormation	0.2297965	0.0397146	5.79	0	0.412381
lnin	-0.1566097	0.3482055	-0.45	0.654	-0.03159
lninitial	-1.481941	0.3892724	-3.81	0	-0.2657
trade	0.0004616	0.0067315	0.07	0.945	0.004951
_IYear_2005	1.903941	0.6739045	2.83	0.005	0.250263
_IYear_2008	-0.93068	0.6784546	-1.37	0.172	-0.1203
_IYear_2011	-0.0143107	0.6918516	-0.02	0.984	-0.00178
_cons	10.91137	3.394956	3.21	0.002	.

How do I report the beta coefficients? Ideally I would like to report it as 'a one standard deviation increase in the Secondary aid decreases growth by x%, How do I do this?

Thankyou
Steph Page

[Dirk Enzmann]

If your dependent variable measures a rate, you should consider using a fractional logit model instead of an OLS regression using glm with a logit link, family binomial, and vce(robust). See http://www.stata.com/meeting/germany...any10_buis.pdf . Using the results you could use margins to predict the change in your dependent variable by a decrease in growth by x% (I cannot comment on the question whether your model makes sense).