Illustrating cross country linear and interaction effects

[Christian BH]

Hello,
I'm trying to illustrate both a linear and an interaction effect across countries using fixed effects, something resembling these two figues from Richardson (2021).

Any ideas how this can be done based on my regression estimates?

Best regards

[Carlo Lazzaro]

Christian:
see -margins- and -marginsplot-.
As an aside, sponsored by the FAQ, full reference, please. Thanks.

[Christian BH]

Hi Carlo,
I creating a marginsplot using the following code:

xtset essround
xtreg c.yvar i.xvar i.controls c.controls i.cntry2, robust fe
margins cntry2
marginsplot, recast(scatter) ytitle(Intolerance) xtitle("") xlabel(, angle(20)) title("") scheme(s1mono) note("95 % confindence interval based on maximum likelihood")

which created this graph where 1) effectsizes corresponding to countries can't be interpreted 2) effects are not ordered after effects:

EDIT: Full reference:
Richardson, Thomas. “Height Is Associated with More Self-Serving Beliefs About Wealth Redistribution.” Evolution and human behavior 42.4 (2021): 287–294. Web.

[Carlo Lazzaro]

Christian:
I do hope the what follows can be helpful:

. use https://www.stata-press.com/data/r17/auto
(1978 automobile data)

. generate tons = weight/2000

. format tons %6.2f

. separate mpg, by(foreign)

Variable      Storage   Display    Value
    name         type    format    label      Variable label
----------------------------------------------------------------------------------------------------------------------------------------
mpg0            byte    %8.0g                 mpg, foreign == Domestic
mpg1            byte    %8.0g                 mpg, foreign == Foreign


. regress mpg foreign##c.tons##c.tons

      Source |       SS           df       MS      Number of obs   =        74
-------------+----------------------------------   F(5, 68)        =     31.82
       Model |   1711.8059         5  342.361181   Prob > F        =    0.0000
    Residual |  731.653555        68  10.7596111   R-squared       =    0.7006
-------------+----------------------------------   Adj R-squared   =    0.6785
       Total |  2443.45946        73  33.4720474   Root MSE        =    3.2802

---------------------------------------------------------------------------------------
                  mpg | Coefficient  Std. err.      t    P>|t|     [95% conf. interval]
----------------------+----------------------------------------------------------------
              foreign |
             Foreign  |   1.592256   23.34741     0.07   0.946    -44.99678    48.18129
                 tons |  -26.34354   9.536229    -2.76   0.007    -45.37279   -7.314293
                      |
       foreign#c.tons |
             Foreign  |  -.0928684   36.88567    -0.00   0.998    -73.69707    73.51133
                      |
        c.tons#c.tons |   4.451742   2.921031     1.52   0.132    -1.377084    10.28057
                      |
foreign#c.tons#c.tons |
             Foreign  |  -2.252604   14.27307    -0.16   0.875    -30.73407    26.22886
                      |
                _cons |    50.7455   7.618435     6.66   0.000     35.54315    65.94786
---------------------------------------------------------------------------------------

. margins, at(tons=(.8(.05)2.4)) over(r.for)

<snip>

. marginsplot, yline(0)

Variables that uniquely identify margins: tons

.

[Christian BH]

If I applied the code correctly the correct code should be (given 33 countries in dataset):

xtset essround
xtreg c.intolerance i.børn i.indkomst c.eduyrs c.agea i.cntry2, robust fe
margins, at(cntry2=(1(1)33)) over(r.børn)
marginsplot, yline(0)

However, this gives me this graph

Best regards

[Carlo Lazzaro]

Christian:
let's change the last line of my previous code as follows:

marginsplot, yline(0) xlabel(, angle(90)) xlabel(#33)

[Christian BH]

Carlo:
It did display all countries but estimates are still wrong

Best regards

[Carlo Lazzaro]

Christian:
what if:

xtset essround
xtreg c.intolerance i.børn i.indkomst c.eduyrs c.agea i.cntry2, robust fe
margins, at(cntry2=(1(1)33))
marginsplot, yline(0) xlabel(,angle(90)) xlabel(#33)

[Christian BH]

This looks more accurate but I the estimates still seem wrong:

For example estimating the effect in Austria (country 1 in dataset) using:

xtset essround
xtreg c.intolerance i.børn c.agea i.indkomst c.eduyrs if cntry2 ==1, robust fe

I get the estimate 0.0168 (insignificant) which doesn't fit with the value >5 on the graph.

Kind regards

[Carlo Lazzaro]

Christian:
please note that the graph reports the linear prediction for each country (that differ from the country-specific coefficients you get).
Just try:

predict fitted, xb

after -xtreg- and compare what you get vs the values reported on the graph.