0
$\begingroup$

Note: Apologies in advance for mistakes in terminology and understanding of the methods, I am a novice at this.

I would like to compare the mental health of different groups of immigrants (by region-of-origin) across several years. The data is from a national survey for a period of 3 years (2016-2018) and my final sample has 7,168 respondents.

I am using 2016 as a reference point, so it seems to me at first glance that pooled cross-sectional data is the way to go with time dummies.

But understanding that there may be problems with unobserved time-invariant, within-region-of-origin effects, should I go for a fixed effects model instead? Would, culture, for example, be considered a time-invariant effect, thus making a fixed effect model more appropriate?

Please correct me if I'm wrong. I am confused, especially since I am not very clear about fixed effects models.

$\endgroup$
4
  • $\begingroup$ Welcome, Lia. By "pooled cross section" you mean you obtain different samples of immigrants in each time period? Is this national survey conducted every year? $\endgroup$ Commented Jan 27, 2021 at 23:16
  • $\begingroup$ Thank you, Thomas. Yes, it is with different samples of immigrants every time. And yes, the survey is conducted every year with nationally representative samples. $\endgroup$
    – Lia
    Commented Jan 27, 2021 at 23:53
  • $\begingroup$ Is it also representative by region-of-origin? Do you observe the same "groups" in each time period? For example, do you observe a sample of approximately 200 Argentines in each time period? Is it possible for a group to be represented in one year but not others? $\endgroup$ Commented Jan 28, 2021 at 0:26
  • $\begingroup$ From my understanding, the proportions of the respondents are representative by region-of-origin. Numerically speaking they do differ, but percentage wise they are close from year to year. Also, all groups have been consistently represented since 2000, so for the past 18 years. To share further, I am using the NHIS data that is available from IPUMS. nhis.ipums.org/nhis $\endgroup$
    – Lia
    Commented Jan 28, 2021 at 12:20

1 Answer 1

1
$\begingroup$

I am using 2016 as a reference point, so it seems to me at first glance that pooled cross-sectional data is the way to go with time dummies.

Perhaps.

It appears your data is more aptly described as repeated cross-sectional data. The survey repeatedly samples a new subset of individuals in each survey wave. You could certainly estimate a standard linear model that 'pools' all available years together, but it will ignore the heterogeneity across groups with ties to different regions of the world.

But understanding that there may be problems with unobserved time-invariant, within-region-of-origin effects, should I go for a fixed effects model instead?

A fixed effects estimation strategy is worth exploring if you suspect the omission of any time-constant confounders that may also be correlated with other explicitly measured variables included in your model. Estimating a "region-of-origin" fixed effect adjusts for all time-invariant heterogeneity, observed and unobserved, across groups from different countries.

Would, culture, for example, be considered a time-invariant effect, thus making a fixed effect model more appropriate?

It depends why you want to treat their country-of-origin as fixed. But, in general, you can do this.

The “culture” (i.e., arts, laws, language, religion, music, diet, etc.) of the immigrant population is likely a time-constant attribute of the individuals that comprise the group. Their learning and socialization process more than likely occurred in their home country, though the ‘acculturation process’ starts to take effect as the group assimilates to the prevailing culture of their host country (e.g., eating the food, learning the language, etc.). Treating their country-of-origin as fixed is one way to adjust for the customs and habits of the group that were ingrained since birth and likely influence the way they interact with others and organize their personal lives. Many of these habits, which will largely be unmeasured, might also influence their emotional/psychological well-being and also be correlated with other variables in your model. Is physical activity part of their daily rituals? How is their diet? Did most individuals attend public school or parochial school, and was it compulsory? Arguably, this is all part of their “culture” before immigrating.

It is difficult to offer further insight without seeing the model under consideration or a subset of your data frame, but I hope this helps.

$\endgroup$
1
  • $\begingroup$ very interesting, I am going through the same problem, a repeated cross sectional data, but with no time related variables, so its fixed effects for me $\endgroup$ Commented Jun 22, 2021 at 1:19

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.