Thanks for the replies, and sorry not to have been clearer and the
delay in responding (student strike!)
the variables are:
label yearsedn "years of education" // cross section in 1999
gen old = born >= 1956 & born <= 1961
gen young = born >- 1970 & born <= 1975
gen highintensity = state >= 5 & state <= 19 // states 1 -4 = low intensity
label highintensity "high primary school funding in 1976-81" // (i.e. treated)
gen young_high = highintensity * young
* Maarten suggests
. reg yearsedn young highintensity young_high if young | old
Source | SS df MS Number of obs = 2698
-------------+------------------------------ F( 3, 2694) = 115.81
Model | 8584.36711 3 2861.4557 Prob > F = 0.0000
Residual | 66562.2808 2694 24.7076024 R-squared = 0.1142
-------------+------------------------------ Adj R-squared = 0.1132
Total | 75146.6479 2697 27.8630508 Root MSE = 4.9707
------------------------------------------------------------------------------
yearsedn | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
young | 2.286417 .4506203 5.07 0.000 1.40282 3.170013
highintens~y | -3.533677 .3945928 -8.96 0.000 -4.307412 -2.759941
young_high | .7222411 .503475 1.43 0.152 -.2649952 1.709477
_cons | 6.118812 .3497354 17.50 0.000 5.433035 6.804589
------------------------------------------------------------------------------
I want a table with mean and standard errors of years of education
where the Xs are in the following table, and the marginal differences
young
0 1 difference
_______________________________________
lo | X | X | X
high | X | X | X
_______________________________________
difference | X | X | X
The row and column Xs can be filled in from calcuations of means (e.g
in ttest), except for the bottom right X
Maarten's suggestion is that the appropriate se is 0.503
Thanks for further help
RIchard
On Tue, Feb 10, 2009 at 5:18 PM, Steven Samuels
<[email protected]> wrote:
> While I agree with Maarten about the analysis procedure, I am unclear about
> the design- does "one of which" in your description refer to region or to
> cohort? Was there a treated cohort and a non-treated cohort within each
> region? Or, were there two cohorts within each region, but with treatment
> differing between region? Do you have a before/after differences , or are
> all comparisons cross-sectional?
>
> Please lay out in more detail the design of the study
>
> -Steve
>
>>
>> --- On Tue, 10/2/09, Richard Palmer-Jones wrote:
>>>
>>> I want to calculate the standard error of the mean
>>> "diffference in difference". I have samples from two
>>> cohorts in each of two regions, one of which received
>>> a treatment, and their respective years of education;
>>> how to I calculate the standard error of the
>>> difference in difference of the years of education?
>>
>> Looks to me like an interaction effect:
>> The dummy for cohort tells you how much the average
>> education differs over cohorts, and you think that
>> this difference differs across regions, so the
>> interaction effect between cohort and region gives
>> you this difference in difference, and -regress- will
>> automatically give you the standard error.
>
> *
> * For searches and help try:
> * http://www.stata.com/help.cgi?search
> * http://www.stata.com/support/statalist/faq
> * http://www.ats.ucla.edu/stat/stata/
>
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
