Search code examples
rplotplmmarginal-effects

Marginal effects plot of PLM

I’ve run an individual-fixed effects panel model in R using the plm-package. I now want to plot the marginal effects. However, neither plot_model() nor effect_plot() work for plm-objects. plot_model() works for type = “est” but not for type = “pred”. My online search so far only suggests using ggplot (which however only displays OLS-regressions, not fixed effects) or outdated functions (i.e, sjp.lm())

Does anyone have any recommendations how I can visualize effects of plm-objects?

IFE_Aut_uc <- plm(LoC_Authorities_rec ~ Compassion_rec, index = c("id","wave"), model = "within", effect = "individual", data = D3_long2)
summary(IFE_Aut_uc)

plot_model(IFE_Aut_uc, type = "pred”)

Error in data.frame(..., check.names = FALSE) : 
  arguments imply differing number of rows: 50238, 82308

and:

effect_plot(IFE_Pol_uc, pred = Compassion_rec)

Error in `stop_wrap()`:
! ~does not appear to be a one- or two-sided formula.
LoC_Politicians_recdoes not appear to be a one- or two-sided formula.
Compassion_recdoes not appear to be a one- or two-sided formula.
Backtrace:
 1. jtools::effect_plot(IFE_Pol_uc, pred = Compassion_rec)
 2. jtools::get_data(model, warn = FALSE)
 4. jtools:::get_lhs(formula)
Solution

  • Edit 2022-08-20: The latest version of plm on CRAN now includes a predict() method for within models. In principle, the commands illustrated below using fixest should now work with plm as well.

    In my experience, plm models are kind of tricky to deal with, and many of the packages which specialize in “post-processing” fail to handle these objects properly.

    One alternative would be to estimate your “within” model using the fixest package and to plot the results using the marginaleffects package. (Disclaimer: I am the marginaleffects author.)

    Note that many of the models estimated by plm are officially supported and tested with marginaleffects (e.g., random effects, Amemiya, Swaymy-Arora). However, this is not the case of this specific "within" model, which is even trickier than the others to support.

    First, we estimate two models to show that the plm and fixest versions are equivalent:

    library(plm)
library(fixest)
library(marginaleffects)
library(modelsummary)
data("EmplUK")

mod1 <- plm(
    emp ~ wage * capital,
    index = c("firm", "year"),
    model = "within",
    effect = "individual",
    data = EmplUK)

mod2 <- feols(
    emp ~ wage * capital | firm,
    se = "standard",
    data = EmplUK)

models <- list("PLM" = mod1, "FIXEST" = mod2)

modelsummary(models)
    PLM FIXEST
    wage 0.000 0.000
    (0.034) (0.034)
    capital 2.014 2.014
    (0.126) (0.126)
    wage × capital -0.043 -0.043
    (0.004) (0.004)
    Num.Obs. 1031 1031
    R2 0.263 0.986
    R2 Adj. 0.145 0.984
    R2 Within 0.263
    R2 Within Adj. 0.260
    AIC 4253.9 4253.9
    BIC 4273.7 4273.7
    RMSE 1.90 1.90
    Std.Errors IID
    FE: firm X

    Now, we use the marginaleffects package to plot the results. There are two main functions for this:

    1. plot_cap(): plot conditional adjusted predictions. How does my predicted outcome change as a function of a covariate?
    2. plot_cme(): plot conditional marginal effects. How does the slope of my model with respect to one variable (i.e., a derivative or “marginal effect”) change with respect to another variable?

    See the website for definitions and details: https://vincentarelbundock.github.io/marginaleffects/

    plot_cap(mod2, condition = "capital")


    
plot_cme(mod2, effect = "wage", condition = "capital")