xtset in cross-sectional data

[Kentaro Kobayashi]

Dear Statalist,

In my research, I am addressing an issue which is called "sibling rivalry".
Sibling rivalry occurs when siblings compete for parental investments.

To account for unobserved family fixed effects common to all siblings, I think that it may be necessary to use a household level fixed effect.
However, my dataset is from a survey conducted in 2017, so the dataset is cross-sectional.

An example of my estimation strategy is the following (since my dataset is from a survey, I use svy: command):

svy: reg log_birthweight rain_shock i.race i.gender i.birth_district i.birth_month

This is the first stage regression where log_birthweight is an endogenous variable and rain_shock is an IV for log_birthweight.

It is ideal for me to add household level fixed effects in this expression, but if I execute the following code,

svy: reg log_birthweight rain_shock i.race i.gender i.birth_district i.birth_month i.hhid

the error

r(103);  too many variables specified

occurs. Since I understand why this happened (because the number of hhid is very large), I thought about taking a different approach which will be explained below.

I recalled that the professor explained in the econometrics class that cross-sectional data can be handled like panel data.
The professor used the following example.

Y_ij = \beta_0 + \beta_1 * X_ij + a_j + u_ij

i: individual, j: school, Y: test score, X: teacher's experience, a: fixed effect, and u: error term.

I thought that this approach could be used in my own setting as well. That is, my idea is using

xtset sibling_id child_id

Is this approach useful in my setting?
Also, if this is true, how should I estimate the 2SLS regression using

svy:

command?

I would greatly appreciate any suggestions you might have.

Best,
Kentaro

[Andrew Musau]

because the number of hhid is very large

On one technical detail, you can absorb the household dummies since there is no requirement to declare a time variable when you xtset your data. Whether your approach in its entirety makes sense, I do not know. Here is an example based on cross-sectional data

sysuse auto
encode make, gen(mk)
set seed 1978
gen car= runiformint(1, 48)
regress mpg price weight i.car
xtset car
xtreg mpg price weight, fe