Difference in difference on a cross sectional data set with small number of cluster group

[Aditi Roy]

Hi,

I am running the following regression: with and without mean group averages in a diff-in-diff cross-sectional data set up.

* Example generated by -dataex-. To install: ssc install dataex
clear
input float(id year_of_birth Post_X_policy years_of_education) str1 states float(sex_ratio literacy_rate Current_age Household_asset Male_head survey_year)
1 1985 1 10 "A" .43 36 35 1 1 2000
2 1986 0 11 "B" .55 21 31 0 1 2005
3 1987 1 15 "A" .46 26 30 1 1 2000
4 1985 1 17 "C" .15 29 25 0 0 2000
5 1991 1 15 "B" .36 24 26 1 1 2005
6 1991 1 14 "A" .66 21 30 0 1 2000
7 1993 0 15 "B" .55 23 22 1 1 2000
8 1985 0 20 "C" .74 55 27 1 0 2005
end

reg years_of_education Post_X_policy sex_ratio literacy_rate Current_age Household_asset Male_head i.states i.year_of_birth, cluster(states)

When I am just doing vce robust I am getting an F test but with clustering my F-test goes missing.

Now my worry of concern is the small clustering group.

My dataset is large but the no of clusters i.e at the state level is low (around 12 or less, whereas the minimum 42clustering group, is the rule of thumb)

How should I deal with this? I am aware of Cameroon's wild bootstrap score approach but as the F test with clustering at the state level is missing. I am afraid.

Given the scenario, should I implement bootstrap clustering rather than going for a state level clustering followed by wild bootstrap score??

[Carlo Lazzaro]

Aditi:
please note:
1) it is impossible to elaborate on your example; please use CODE delimiters to post. Thanks;
2) missing F-test is well covered in help j_robustsingular (in all likelihhod, you have a freedom degrees problem due to clustered standard errors);
3) as often reminded on this list, urgency is poster's own business.