texreg-ing multiple multinom models

问题

I'm having a bit of trouble using texreg to create output TeX files for multiple multinom models.

Let's use a version of this multionmial logit setup for concreteness:

library(foreign)
library(nnet)
library(data.table)
ml <- data.table(read.dta("http://www.ats.ucla.edu/stat/data/hsbdemo.dta"))

mnl<-lapply(c(model1="male",model2="female"),
            function(x){multinom(prog ~ ses + write,
                                 data = ml[female==x,])})

I want to create two tables: one for coefficients corresponding to "academic" and one corresponding to "vocation" (the third outcome of prog, "general", is omitted).

Basically, I don't know how separately access the levels from an nnet/multinom object, because I don't know where the coefficients are stored. If we just run texreg(mnl), we'll get a table with 4 columns:

model1 & model2 & NA & NA \\

What's worse, these columns are mislabeled--the actual order is

model1-academic, model1-vocation, model2-academic, model2-vocation

If I simply wanted a table of male and a table of female coefficients, I would simply run texreg(mnl["model1"]) and texreg(mnl["model2"]), but it's not clear how to split the coefficients by levels of prog.

How can I use texreg to get the two tables I want, which (skeletally) look like:

#TABLE 1: ACADEMIC
> cbind(male=coef(mnl[["model1"]])[1,],
+       female=coef(mnl[["model2"]])[1,])
                   male      female
(Intercept) -3.23410968 -3.01401061
sesmiddle    1.15893835  0.17086744
seshigh      2.00007946  0.57690699
write        0.05464579  0.06598217

#TABLE 2: VOCATION
> cbind(male=coef(mnl[["model1"]])[2,],
+       female=coef(mnl[["model2"]])[2,])
                   male      female
(Intercept)  3.69215046  1.57234796
sesmiddle    1.15573930  0.69043245
seshigh      0.67476976 -0.16955825
write       -0.09640053 -0.03412729

回答1:

Got in contact with package developer Prof. Philip Leifeld, and he so graciously added functionality for handling exactly this problem to texreg, specifically adding two options: levels and beside.

The answer to my original question is accomplished through levels:

texreg(mnl,levels="vocation")
texreg(mnl,levels="academic")

\begin{table}
\begin{center}
\begin{tabular}{l c c }
\hline
                & vocation & vocation \\
\hline
(Intercept)     & $3.69^{*}$  & $1.57$   \\
                & $(1.77)$    & $(1.84)$ \\
sesmiddle       & $1.16$      & $0.69$   \\
                & $(0.79)$    & $(0.65)$ \\
seshigh         & $0.67$      & $-0.17$  \\
                & $(0.99)$    & $(0.91)$ \\
write           & $-0.10^{*}$ & $-0.03$  \\
                & $(0.04)$    & $(0.04)$ \\
\hline
AIC             & 167.56      & 218.73   \\
BIC             & 187.65      & 240.26   \\
Log\ Likelihood & -75.78      & -101.36  \\
Deviance        & 151.56      & 202.73   \\
Num.\ obs.      & 91          & 109      \\
\hline
\multicolumn{3}{l}{\scriptsize{$^{***}p<0.001$, $^{**}p<0.01$, $^*p<0.05$}}
\end{tabular}
\caption{Statistical models}
\label{table:coefficients}
\end{center}
\end{table}

(I'll suppress output from "academic" to save space, but you get the point).

The beside functionality works slightly differently, instead stacking the coefficients, if that's what suits your fancy:

\begin{table}
\begin{center}
\begin{tabular}{l c c }
\hline
                      & model1 & model2 \\
\hline
academic: (Intercept) & $-3.23$     & $-3.01$    \\
                      & $(1.66)$    & $(1.82)$   \\
academic: sesmiddle   & $1.16$      & $0.17$     \\
                      & $(0.75)$    & $(0.58)$   \\
academic: seshigh     & $2.00^{*}$  & $0.58$     \\
                      & $(0.85)$    & $(0.69)$   \\
academic: write       & $0.05$      & $0.07^{*}$ \\
                      & $(0.03)$    & $(0.03)$   \\
vocation: (Intercept) & $3.69^{*}$  & $1.57$     \\
                      & $(1.77)$    & $(1.84)$   \\
vocation: sesmiddle   & $1.16$      & $0.69$     \\
                      & $(0.79)$    & $(0.65)$   \\
vocation: seshigh     & $0.67$      & $-0.17$    \\
                      & $(0.99)$    & $(0.91)$   \\
vocation: write       & $-0.10^{*}$ & $-0.03$    \\
                      & $(0.04)$    & $(0.04)$   \\
\hline
AIC                   & 167.56      & 218.73     \\
BIC                   & 187.65      & 240.26     \\
Log\ Likelihood       & -75.78      & -101.36    \\
Deviance              & 151.56      & 202.73     \\
Num.\ obs.            & 91          & 109        \\
\hline
\multicolumn{3}{l}{\scriptsize{$^{***}p<0.001$, $^{**}p<0.01$, $^*p<0.05$}}
\end{tabular}
\caption{Statistical models}
\label{table:coefficients}
\end{center}
\end{table}

来源：https://stackoverflow.com/questions/30017129/texreg-ing-multiple-multinom-models

标签

texreg