Dears,
I'm using the user-written command mkspline2 to estimate restricted cubic splines (by Maarten Buis). The dependent variable is ostck (stock ownership) and the regressor of interest is continuous (lexemp1). I am estimating models with 4 knots, using individual and state fixed-effects as follows:
mkspline2 lexx = lexemp1, cubic nknots(4) displayknots
xtreg ostck lexx1 lexx2 lexx3 demographiccontrols statetrend* i.state i.wave,fe vce(cluster state) nonest
And I get this outcome for the spline parameters (I'm only copying the regressors of interest here to save space):
--------------------------------------------------------------------------------------------------------
| Robust
ostck | Coef. Std. Err. P>|t| [95% Conf. Interval]
-------------+------------------------------------------------------------------------------------------
lexx1 | .0161402 .0134662 1.20 0.236 -.0109076 .0431879
lexx2 | -.2020433 .0819126 -2.47 0.017 -.3665696 -.037517
lexx3 | .5007041 .196728 2.55 0.014 .1055643 .895844
And then I plot a graph for the adjusted predictions using adjustrcspline (part of the same package):
adjustrcspline, link(identity) recast(line) yline(0) xline(9.012874 9.975579 11.1425 13.24698, lpattern(dot)) ciopts(color(gold*.8)) level(90) title(" " "Adjusted Predictions with 90% CIs") xtitle(lexemp1) ytitle(Pr(Own Stock))
And a graph for the marginal effects using mfxrcspline (also in the same package):
mfxrcspline, link(identity) recast(line) yline(0) xline(9.012874 9.975579 11.1425 13.24698, lpattern(dot)) ciopts(color(*.4)) level(90) title(" " "Conditional Marginal Effects with 90% CIs") xtitle(lexemp1) ytitle(d[Pr(Own Stock)]/d(lexemp1))
I am attaching the graphs that I obtain. When I look at the adjusted margins, based on the confidence intervals, it seems as if the effects were not significant, whereas the graph for the marginal effects suggest that the effects are significantly negative at the interval 10-11 of lexemp1 (and also the coefficients of the nonlinear terms of the regression are significant). Assuming that I am using the commands correctly, I wonder if there is something about the interpretation of the graphs (particularly for the adjusted predictions) that I am not taking into account.
Thanks a lot!
Mariela
I'm using the user-written command mkspline2 to estimate restricted cubic splines (by Maarten Buis). The dependent variable is ostck (stock ownership) and the regressor of interest is continuous (lexemp1). I am estimating models with 4 knots, using individual and state fixed-effects as follows:
mkspline2 lexx = lexemp1, cubic nknots(4) displayknots
xtreg ostck lexx1 lexx2 lexx3 demographiccontrols statetrend* i.state i.wave,fe vce(cluster state) nonest
And I get this outcome for the spline parameters (I'm only copying the regressors of interest here to save space):
--------------------------------------------------------------------------------------------------------
| Robust
ostck | Coef. Std. Err. P>|t| [95% Conf. Interval]
-------------+------------------------------------------------------------------------------------------
lexx1 | .0161402 .0134662 1.20 0.236 -.0109076 .0431879
lexx2 | -.2020433 .0819126 -2.47 0.017 -.3665696 -.037517
lexx3 | .5007041 .196728 2.55 0.014 .1055643 .895844
And then I plot a graph for the adjusted predictions using adjustrcspline (part of the same package):
adjustrcspline, link(identity) recast(line) yline(0) xline(9.012874 9.975579 11.1425 13.24698, lpattern(dot)) ciopts(color(gold*.8)) level(90) title(" " "Adjusted Predictions with 90% CIs") xtitle(lexemp1) ytitle(Pr(Own Stock))
And a graph for the marginal effects using mfxrcspline (also in the same package):
mfxrcspline, link(identity) recast(line) yline(0) xline(9.012874 9.975579 11.1425 13.24698, lpattern(dot)) ciopts(color(*.4)) level(90) title(" " "Conditional Marginal Effects with 90% CIs") xtitle(lexemp1) ytitle(d[Pr(Own Stock)]/d(lexemp1))
I am attaching the graphs that I obtain. When I look at the adjusted margins, based on the confidence intervals, it seems as if the effects were not significant, whereas the graph for the marginal effects suggest that the effects are significantly negative at the interval 10-11 of lexemp1 (and also the coefficients of the nonlinear terms of the regression are significant). Assuming that I am using the commands correctly, I wonder if there is something about the interpretation of the graphs (particularly for the adjusted predictions) that I am not taking into account.
Thanks a lot!
Mariela
Attached Files
