Fixed effects for IV using cross section data

[Akanksha Soni]

Dear Statlist users,

I am studying the impact of parental investments in early childhood, as recorded in the 1993 data set on adult labour market earnings(as observed in 2005). While the data set is household level panel data, what I am using is cross section as the investments in 1993 are not observed in 2005, just the impact on income in 2005. I have used areg to do a siblings fixed effects analysis, however I am struggling with doing an IV with sibling fixed effects. I wish to use the command xi: ivreg2 depvar (endogenous var=instrument) i.var, however this I understand i.var can be used only for categorical variables. My data is a survey data that contains mother's id that takes the value 1-28 and would take the same value for siblings. STATA therefore gives me a result like this for each of the 28 values:

parentaliindex | -3225.741 542.0528 -5.95 0.000 -4288.145 -2163.337
_Imothersid_1 | -7207.049 4750.07 -1.52 0.129 -16517.02 2102.917
_Imothersid_2 | -1449.11 3207.187 -0.45 0.651 -7735.082 4836.862
_Imothersid_3 | 14257.66 3713.269 3.84 0.000 6979.788 21535.53
_Imothersid_4 | 19519.84 3493.853 5.59 0.000 12672.01 26367.67
_Imothersid_5 | 33893.13 4362.288 7.77 0.000 25343.2 42443.06
_Imothersid_6 | 37477.25 4546.112 8.24 0.000 28567.04 46387.47
_Imothersid_7 | 29743.94 4651.685 6.39 0.000 20626.8 38861.07
_Imothersid_8 | 34198.41 4574.421 7.48 0.000 25232.71 43164.11

Is there a way I can do an IV for cross section data whilst including siblings/family fixed effects?

Regards,
Akanksha

[Jeff Wooldridge]

Is it true there are only 28 mothers in the sample? That would be pretty small for doing IV.

[Akanksha Soni]

Dear Jeff Wooldridge

It is a survey data set so mother's id takes repeated values of 1-28. There are around 60,000 observations.

[Jeff Wooldridge]

That seems to be a poor convention. How can the different mothers be identified?

[Jeff Wooldridge]

That seems to be a poor convention. How can the different mothers be identified?

[Akanksha Soni]

There is no unique id for the mother. So you can generate one, using a concatenation of state code, district code, psu code, household id.

[Jeff Wooldridge]

Have you generated the unique id for the mothers? If so, you can get xtivreg to do the estimation by using

xtset motherid

[Akanksha Soni]

Yes I have generated an ID however, my data set is technically not panel as the treatment is observed in 1993 and the outcome in 2005. That is the reason I am unable to declare it a panel using xtset.
For my fixed effects analysis I used areg indep depvar, absorb(household/motherid)
Is there something similar I can do for an IV estimation using fixed effects?

[Jeff Wooldridge]

I know it’s not true panel data but xtivreg doesn’t actually care. Try this:

xtset motherid
xtivreg y x1... xk (w = z1 ... zm), fe vce(cluster, motherid)

[Akanksha Soni]

I tried that but I get an error message on STATA that says this:

xtivreg INCOME (parentaliindex=instrument), fe vce(cluster, idhh)
the sample specifies cross-sectional data
xtivreg is not designed for cross-sectional data
use ivregress with cross-sectional data
r(498);

I was wondering if ivreghdfe Y x1............xk (x=z), absorb(mothersid) would be the right command? As it allows for multilevel fixed effects?

[Akanksha Soni]

I get unusually high p values with ivreghdfe and fail to understand the results

[Jeff Wooldridge]

Birth order and age clearly vary by child. But what is the parental index variable?

I got xtivreg to work in a similar setting so I’m not sure why it doesn’t in yours.