Interpretation of education dummies in mincer equation

[Sayoree Gooptu]

I am using a mincerian wage equation to estimate the returns to education. I came across two ways to include the education dummies.

i) Using the normal dummy variable command by i.education and regressing log wage on that after including other controls. The base category is taken as illiterates
ii) In the other method, I have used the sequential dummies. For instance, a for a graduate, the dummies for education levels below graduation takes the value 1 and for higher than graduation degree it takes value zero

gen educ=.
replace educ=1 if General_edulevel<=4 & General_edulevel>=1
replace educ=2 if General_edulevel==5
replace educ=3 if General_edulevel==6
replace educ=4 if General_edulevel==7
replace educ=5 if General_edulevel==8
replace educ=6 if General_edulevel==10|General_edulevel==11
replace educ=7 if General_edulevel==12|General_edulevel==13
label define levedu 1 "Not Literate" 2 "Below Primary" 3 "Primary" 4 "middle" 5 "secondary" 6 "Higher Secondary"  7 "Graduate & above"
label values educ levedu

qui tab educ, gen (ed)
levelsof educ, local(levs)
foreach l of local levs{
replace ed`l'=1 if educ>=`l'
}

I am a bit confused on which method to use to calculate the returns to education for each education level. Any suggestion on this would be extremely helpful.

[Carlo Lazzaro]

Sayoree:
I would go with i), keeping in mind that your regression might be affected by latent variable-led endogeneity (individual's ability is the issue here; see, among many other contributions; https://arxiv.org/ftp/arxiv/papers/1711/1711.05076.pdf; page 3, d)): on average, other things being equal, smarter persons obtain higher education levels; on average, other things being equal, smarter persons negotiate higher wages.

[Sayoree Gooptu]

Carlo Lazzaro Thanks a lot for the reply.
Actually my main research question is to estimate the extent of overeducation in certain occupations. I use the Mincer equation for that. I consider an education level to be appropriate level of education, if the returns are significant. But if I go by the first method, the coefficients reflect the wage penalty/premium compared to illiterates. Since I require the returns for each education, I was thinking of the second approach. I do not want to compare with the wages of illiterates, but just the returns to each level of education. Wouldn't the latent variable led endogeneity be possibly addressed in this method as well?

Also, could please share the link of the paper a second time? The page seems to be unavailable.

[Carlo Lazzaro]

Sayoree:
1) the title of the paper is: The return to higher education: evidence from Romania (sorry, but I'm away from my PC at the moment and cannot repost the link that is actually broken);
2) I fail to get how individual's ability can be addressed via the second approach unless you instrument it appropriately.

[Sayoree Gooptu]

Carlo Lazzaro Thanks again for your kind response.

So in the second approach, I consider two people, one with secondary education another with graduation. The dummy variables are as follows

primary middle secondary highersec graduation
1 1 1 0 0
1 1 1 1 1

Doesn't that also capture that higher education levels indicate higher ability? A person with 1 as graduation dummy must have 1 for all the levels below. In that sense, it is also indicating a higher level of ability? The only difference lies in interpretation of the dummies, to which level are we comparing the coefficients. I was confused on that. I may be wrong with my understanding. Actually I particularly need the returns to each level and to find whether they are significant or not. So I was following this method. It is similar to a paper by Mehta et al. 2011, https://www.sciencedirect.com/scienc...038?via%3Dihub . Please correct me if I am wrong. It would be really helpful.

[Carlo Lazzaro]

Sayoree:
the paper you mentioned explore overeducation.
In thar setting, it's hard to believe in latent variable-led endogeneity, as the research focuses on increasing education levels in unskilled jobs.
Therefore: are you regressing wage on education or overeducation (+ other predictors and controls)?

[Sayoree Gooptu]

Carlo Lazzaro

I am regressing wages on the education levels (+other predictors and controls). I have taken the specification following Mehta et al. (2011). There they have considered a job to be unskilled if there is no significant returns to secondary and tertiary education. After that, they have considered the proportion of secondary and tertiary employed in those jobs as overeducated. The education dummies they have regressed wage on is the one I described in my previous post.

[Carlo Lazzaro]

Sayoree:
roger that.
I've nothing more to add to my previous comments.

[Sayoree Gooptu]

Carlo Lazzaro

Okay.

Thanks a lot for your kind and helpful response