Does this regression looks linear?

[Pedro Martinez]

Hello, does this regression looks linear? If not what are the steps to solve it ? Thanks

[Dario Maimone Ansaldo Patti]

Hi,

you mean linear in the regressors you are using i guess. Well, in my view it is difficult to answer your question. You have many observations in your dataset. Nonetheless, i would not make a decision about linearity/not linearit, based on tma graph. First i would check the relevant literature to see if it offers any suggestion. Eventually, if not linearity can be justified on a theoretical ground, then i would check if your data fits within a not linear framework. There are some specific tests that you can run, for instance utest (ssc install utest) to check for a quadratic U shape or inverse U shape relationship.

Looking at a graph can provide a first preliminary intuition but not a final word about the specification you should use

[Carlo Lazzaro]

Pedro:
as an aside to Dario's helpful advice, I would add:
1) interested listers cannot be sure of what you've plotted on your graph. Is it fitted values against a given predictor?
2) have you searched for possible turning points in your regression equation?
3) have you tested that the functional form of the regressand is correctly specified (see -linktest-)?
4) have you tested for heteroskedastcity in the residual distribution and, in case you detected it, have you considered transforming the regressand or imposing -robust- standard errors?

[Felix Bittmann]

For the graph I suggest using binscatter

ssc install binscatter, replace

https://michaelstepner.com/binscatter/

[Nick Cox]

I suggest using lpoly. Further, for your graph, large solid circles are the worst possible choice. See below for a better one.

Reproducible example:

webuse nlswork, clear

scatter ln_wage birth_yr , ms(oh) mc(blue%10) || lpoly ln_wage birth_yr, degree(1)