问题
I am new to R, and I don't understand yet completely the logic of its calculations...
I cannot overcome my problem with the help from previous posts either.
I have a data set of about 600 observations for 11 variables. I have successfully run the multinomial model on it, however I cannot achieve the marginal effects because my mean() command is getting NAs:
The data set:
> head(data,n=50)
ID time CHINN DEBT ERA INFL MONEY OPENNESS RESERVES RGDP RSVS
1 POL 1993 -1.8639720 NA 0 32.8815343 33.47353 41.05223 4276726138 100.00000 4.2767261
2 POL 1994 -0.8081098 NA 0 30.7586977 31.98628 41.26984 6023061884 105.13912 6.0230619
3 POL 1995 -0.1129656 NA 0 26.0622777 31.69277 44.24652 14957024390 112.15091 14.9570244
4 POL 1996 -1.1688280 NA 0 19.1794515 33.19738 46.04792 18018686050 119.00579 18.0186860
5 POL 1997 -1.1688280 NA 0 14.8181395 35.11881 50.74269 20669498248 127.43530 20.6694982
6 POL 1998 -1.1688280 NA 4 11.1353669 37.70208 56.79412 28276350644 133.65597 28.2763506
7 POL 1999 -1.1688280 NA 4 7.0830285 40.69941 54.29511 27314254555 139.74225 27.3142546
8 POL 2000 -1.1688280 32.96343 1 10.7321436 40.60334 60.67079 27469379821 145.66606 27.4693798
9 POL 2001 -1.1688280 41.37857 1 4.3657643 44.43624 57.77678 26563086922 147.25896 26.5630869
10 POL 2002 0.0644257 NA 1 0.4689145 41.98616 60.73279 29783861006 149.32464 29.7838610
11 POL 2003 0.0644257 NA 1 -0.3092132 42.42629 69.31849 33958963841 154.95895 33.9589638
12 POL 2004 0.0644257 NA 1 2.5056441 40.21144 77.33549 36772764782 162.98411 36.7727648
13 POL 2005 0.0644257 NA 1 1.4888388 43.50525 74.91127 42560657450 168.54956 42.5606575
14 POL 2006 0.0644257 NA 1 0.8029315 46.92580 82.51674 48473947849 178.83656 48.4739478
15 POL 2007 0.0644257 NA 1 0.7577241 47.82156 84.38853 65724834811 190.53326 65.7248348
16 POL 2008 0.0644257 NA 1 3.5756117 52.36352 83.76331 62183606786 199.83465 62.1836068
17 POL 2009 0.0644257 NA 1 2.6459580 53.71374 78.80573 79521598778 203.40385 79.5215988
18 POL 2010 0.0644257 NA 1 1.6733804 55.42878 85.68774 93472496388 211.50103 93.4724964
19 POL 2011 0.0644257 NA 1 3.0274845 57.87980 91.29239 97712443397 220.71189 97.7124434
20 BGR 1993 NA NA 1 68.8945207 78.96788 84.04149 1052450357 100.00000 1.0524504
21 BGR 1994 -1.1688280 NA 1 93.5639075 79.70282 90.72013 1396927258 101.92207 1.3969273
22 BGR 1995 -1.1688280 NA 1 60.0454149 67.02978 101.82717 1635188166 104.48391 1.6351882
23 BGR 1996 -0.9050840 NA 1 120.9697303 83.30492 116.19712 864262494 95.38304 0.8642625
24 BGR 1997 -0.9050840 NA 5 1058.1103740 35.17444 112.09386 2485359931 93.68869 2.4853599
25 BGR 1998 -0.9050840 NA 6 18.0824730 28.90920 117.17808 3056954172 97.96299 3.0569542
26 BGR 1999 -0.9050840 NA 6 2.3810375 30.99876 116.55936 3264673405 99.87455 3.2646734
27 BGR 2000 -0.9050840 NA 6 10.9885883 35.97481 106.26211 3507199103 105.59192 3.5071991
28 BGR 2001 -1.1688280 NA 6 6.2354669 40.92721 106.86605 3646131855 109.94781 3.6461319
29 BGR 2002 -1.1688280 NA 6 4.3788710 41.75304 102.98033 4846429828 114.94803 4.8464298
30 BGR 2003 -0.9050840 NA 6 1.0599753 46.25984 107.38463 6825720096 120.93287 6.8257201
31 BGR 2004 -0.6413398 NA 6 5.2752300 51.21173 115.32294 9337108247 128.76534 9.3371082
32 BGR 2005 -0.3775955 NA 6 4.4206316 55.53535 96.16404 8697081229 136.82237 8.6970812
33 BGR 2006 2.1752650 NA 6 6.9496009 61.91483 140.00543 11756001804 145.51154 11.7560018
34 BGR 2007 2.4390090 NA 6 6.7721512 69.93833 138.64999 17544560083 154.43412 17.5445601
35 BGR 2008 2.4390090 NA 6 11.5750056 66.12314 136.94939 17930378450 163.65291 17.9303784
36 BGR 2009 2.4390090 13.82140 6 1.5731668 69.86327 103.84837 18522120691 154.75483 18.5221207
37 BGR 2010 2.4390090 14.93887 6 1.4049372 71.96152 116.71516 17223201500 155.40066 17.2232015
38 BGR 2011 2.4390090 15.44620 6 2.9890214 75.58550 132.27417 17215734344 158.00846 17.2157343
39 CYP 1993 -0.1129656 NA 5 0.8698960 129.36758 95.41637 1276350873 100.00000 1.2763509
40 CYP 1994 -0.1129656 NA 5 2.2051585 131.48065 95.77615 1640273818 105.77324 1.6402738
41 CYP 1995 -0.1129656 NA 5 0.6063991 128.64850 100.10371 1294935266 111.96553 1.2949353
42 CYP 1996 -1.1688280 NA 5 2.3411552 136.94871 104.44163 1704484151 114.02390 1.7044842
43 CYP 1997 -1.1688280 NA 5 3.3418671 145.38644 105.39959 1525648570 116.75559 1.5256486
44 CYP 1998 -1.1688280 261.72872 4 1.6379206 147.00670 100.52824 1512684182 122.55893 1.5126842
45 CYP 1999 -1.1688280 155.75745 4 1.4380285 159.30150 101.75789 1967461070 128.47057 1.9674611
46 CYP 2000 -1.1688280 156.25023 4 4.8139861 161.29472 109.87129 1868515552 135.00970 1.8685156
47 CYP 2001 -1.1688280 160.87807 4 0.8515728 169.52762 109.90206 2396061109 140.39178 2.3960611
48 CYP 2002 -1.1688280 170.52011 4 1.3698956 175.62242 103.06500 3181358048 143.20035 3.1813580
49 CYP 2003 -0.1129656 179.11164 4 3.0419336 174.50803 95.17382 3450932640 145.59528 3.4509326
50 CYP 2004 1.3840320 180.22661 4 1.2153140 171.42676 98.02155 4113824175 151.47324 4.1138242
commands for transforming the data:
pdata<-plm.data(data,index=c("ID","time"))
mldata<-mlogit.data(pdata,choice="ERA",shape="wide")
command for the matrix of means values:
z<-with(mldata,data.frame(CHINN=mean(CHINN),DEBT=mean(DEBT),INFL=mean(INFL),MONEY=mean(MONEY),OPENNESS=mean(OPENNESS),RGDP=mean(RGDP),RSVS=mean(RSVS)))
The output I get:
CHINN DEBT INFL MONEY OPENNESS RGDP RSVS
1 NA NA NA NA NA 133.4633 NA
Could you advise on the reason of the NAs for this command?
I would understand if it would not give a correct output for DEBT which is mostly NAs, but why it is not calculating the mean for RSVS, CHINN and others?
Why does it calculate the mean for RGDP? Both variables has:
> class(mldata$CHINN)
[1] "numeric"
> class(mldata$RGDP)
[1] "numeric"
How to overcome this problem?
UPDATE:
Thanks to the David Arenburg's comment I managed to calculate the means:
> z
CHINN DEBT INFL MONEY OPENNESS RGDP RSVS
1 1.342326 59.85562 33.88494 58.2304 95.88219 133.4633 29.21734
However there is another error that occurs while calculating the marginal effects:
effects(mlogit.data2,covariate="CHINN",data=z)
Error in predict.mlogit(object, data) :
the number of rows of the data.frame should be a multiple of the number of alternatives
After copying the means 5 times to get a matrix zz of 6 rows for 6 alternatives:
> zz
CHINN DEBT INFL MONEY OPENNESS RGDP RSVS
1 1.342326 59.85562 33.88494 58.2304 95.88219 133.4633 29.21734
2 1.342326 59.85562 33.88494 58.2304 95.88219 133.4633 29.21734
3 1.342326 59.85562 33.88494 58.2304 95.88219 133.4633 29.21734
4 1.342326 59.85562 33.88494 58.2304 95.88219 133.4633 29.21734
5 1.342326 59.85562 33.88494 58.2304 95.88219 133.4633 29.21734
6 1.342326 59.85562 33.88494 58.2304 95.88219 133.4633 29.21734
The effects() command gives an output:
> effects(mlogit.data2,covariate=c("CHINN","DEBT","INFL","MONEY","OPENNESS","RGDP","RSVS"),data=zz)
2 0 1 4 5 6
-1.288185e-02 -1.933325e-02 -8.801676e-06 1.095252e-02 2.120814e-02 6.324152e-05
However I cannot draw any conclusions on this output - I do not know to which variable I can assign those marginal effects.
When I try running one by one I get another error:
> effects(mlogit.data2,covariate=c("CHINN"),data=zz)
Error in if (rhs %in% c(1, 3)) { : argument is of length zero
The model command:
mlogit.data2<-mlogit(ERA~1|CHINN+INFL+MONEY+OPENNESS+RGDP+RSVS+DEBT,data=mldata,reflevel="4")
The packages I am using:
library(Formula)
library(miscTools)
library(lattice)
library(zoo)
library(sandwich)
library(maxLik)
library(lmtest)
library(statmod)
library(mlogit)
library(plm)
Thank you in advance, Zyta
回答1:
Ok, seems like I have kind of solution with the help of:
http://www.talkstats.com/showthread.php/44314-calculate-marginal-effects-using-mlogit-package
and
How can I view the source code for a function?
apparently it was enough to see for the:
> getAnywhere(effects.mlogit)
A single object matching ‘effects.mlogit’ was found
It was found in the following places
registered S3 method for effects from namespace mlogit
namespace:mlogit
with value
function (object, covariate = NULL, type = c("aa", "ar", "rr",
"ra"), data = NULL, ...)
{
type <- match.arg(type)
if (is.null(data)) {
P <- predict(object, returnData = TRUE)
data <- attr(P, "data")
attr(P, "data") <- NULL
}
else P <- predict(object, data)
newdata <- data
J <- length(P)
alt.levels <- names(P)
pVar <- substr(type, 1, 1)
xVar <- substr(type, 2, 2)
cov.list <- lapply(attr(formula(object), "rhs"), as.character)
rhs <- sapply(cov.list, function(x) length(na.omit(match(x,
covariate))) > 0)
rhs <- (1:length(cov.list))[rhs]
eps <- 1e-05
if (rhs %in% c(1, 3)) {
if (rhs == 3) {
theCoef <- paste(alt.levels, covariate, sep = ":")
theCoef <- coef(object)[theCoef]
}
else theCoef <- coef(object)[covariate]
me <- c()
for (l in 1:J) {
newdata[l, covariate] <- data[l, covariate] + eps
newP <- predict(object, newdata)
me <- rbind(me, (newP - P)/eps)
newdata <- data
}
if (pVar == "r")
me <- t(t(me)/P)
if (xVar == "r")
me <- me * matrix(rep(data[[covariate]], J), J)
dimnames(me) <- list(alt.levels, alt.levels)
}
if (rhs == 2) {
newdata[, covariate] <- data[, covariate] + eps
newP <- predict(object, newdata)
me <- (newP - P)/eps
if (pVar == "r")
me <- me/P
if (xVar == "r")
me <- me * data[[covariate]]
names(me) <- alt.levels
}
me
}
<environment: namespace:mlogit>
Then to copy the function and modyfy its 16 line:
myeffects<-function (object, covariate = NULL, type = c("aa", "ar", "rr",
"ra"), data = NULL, ...)
{
type <- match.arg(type)
if (is.null(data)) {
P <- predict(object, returnData = TRUE)
data <- attr(P, "data")
attr(P, "data") <- NULL
}
else P <- predict(object, data)
newdata <- data
J <- length(P)
alt.levels <- names(P)
pVar <- substr(type, 1, 1)
xVar <- substr(type, 2, 2)
cov.list <- strsplit(as.character(attr(formula(object), "rhs")), " + ", fixed = TRUE)
rhs <- sapply(cov.list, function(x) length(na.omit(match(x,
covariate))) > 0)
rhs <- (1:length(cov.list))[rhs]
eps <- 1e-05
if (rhs %in% c(1, 3)) {
if (rhs == 3) {
theCoef <- paste(alt.levels, covariate, sep = ":")
theCoef <- coef(object)[theCoef]
}
else theCoef <- coef(object)[covariate]
me <- c()
for (l in 1:J) {
newdata[l, covariate] <- data[l, covariate] + eps
newP <- predict(object, newdata)
me <- rbind(me, (newP - P)/eps)
newdata <- data
}
if (pVar == "r")
me <- t(t(me)/P)
if (xVar == "r")
me <- me * matrix(rep(data[[covariate]], J), J)
dimnames(me) <- list(alt.levels, alt.levels)
}
if (rhs == 2) {
newdata[, covariate] <- data[, covariate] + eps
newP <- predict(object, newdata)
me <- (newP - P)/eps
if (pVar == "r")
me <- me/P
if (xVar == "r")
me <- me * data[[covariate]]
names(me) <- alt.levels
}
me
}
Now the results are as of:
> myeffects(mlogit.data2,covariate="RSVS",data=zz)
2 0 1 4 5 6
3.612318e-03 5.368693e-04 -4.903995e-08 -5.382731e-03 1.238768e-03 -5.175053e-06
回答2:
You could use colMeans
op <- options(scipen= 100, digits=2)
colMeans(mldata[,3:11], na.rm=TRUE)
# CHINN DEBT ERA INFL MONEY
# -0.27 115.25 0.20 33.66 74.66
#OPENNESS RESERVES RGDP RSVS
# 92.06 18809465124.14 136.56 18.81
options(op)
or summarise_each from dplyr
library(dplyr)
mldata %>%
summarise_each(funs(round(mean(., na.rm=TRUE),2)), CHINN:RSVS)
# CHINN DEBT ERA INFL MONEY OPENNESS RESERVES RGDP RSVS
#1 -0.27 115.25 0.2 33.66 74.66 92.06 18809465124 136.56 18.81
来源：https://stackoverflow.com/questions/25831729/marginal-effects-of-mlogit-in-r