R: Ordered logistic regression with multiple imputation data (amelia package)

Question

I am analyzing data from European Social Survey. Due to quite a bit of missing data I have used the amelia package for imputation. The dependent value is ordinal with 4 values, and I had therefore planned to perform a ordered logistic regression with the "ologit" function in the Zelig-package:

z.out <- zelig(as.factor(Y) ~ X1 + X2, model = "ologit", data = ameliadata)

That code will run, but when I ask for the results the following error code is shown:

z.out:

Model: Combined Imputations Error in se[i, ] <- sqrt(diag(vcovlist[[i]])) : number of items to replace is not a multiple of replacement length

I have five separate imputed datasets. Analyzed separately I am able to use Zelig and the "ologit"-function with each of these five. The problem only arises when I use my combined amelia data-object. I have tried to estimate different models with the same amelia-output, and I only seem to have a problem with the ones related to ordered regression. For example, the "ls"-model runs just fine, and if I change the depended variable to a dichotomous I can also run a "logit"-model without problems.

I am therefore wondering whether anyone has been able to run "ologit" with zelig on amelia data previously or if anyone has any idea about what could be the problem? I will greatly appreciate any ideas and suggestions. Thank you so much for your time and help.

This is an example with the wine dataset from the ordinal package:

library(Amelia)
library(Zelig)
library(ordinal)

data(wine)
w <- wine

set.seed(10)
w[sample(1:nrow(w), 20), "response"] <- NA
w[sample(1:nrow(w), 20), "rating"] <- NA
w[sample(1:nrow(w), 20), "temp"] <- NA
w[sample(1:nrow(w), 5), "contact"] <- NA
w[sample(1:nrow(w), 5), "bottle"] <- NA


w.amelia <- amelia(w, m = 5, idvars="bottle", ords = c("rating","judge"),
                     noms = c("contact", "temp"),
                     incheck = TRUE)

z.out <- zelig(rating ~ contact + temp, model = "ologit", data = w.amelia)

summary(z.out)

claudius claudius · Accepted Answer · 2019-09-23T14:57:02

Looks like that the zilig function (with model = "ologit") is not working well with the amelia object.
So to do that you could call the zilig function individually to each one of the 5 imputed dataset with the amelia package. Below we can see the fitted models for two imputed dataset.

z.out1 <- zelig(rating ~ contact + temp, model = "ologit", data = w.amelia$imputations$imp1)
z.out2 <- zelig(rating ~ contact + temp, model = "ologit", data = w.amelia$imputations$imp2)

Receiving an output to each imputed data:

> summary(z.out1)
Model: 
Call:
z5$zelig(formula = rating ~ contact + temp, data = w.amelia$imputations$imp1)

Coefficients:
           Value Std. Error t value
contactyes 1.973     0.4937   3.997
tempwarm   1.493     0.4617   3.235

Intercepts:
    Value   Std. Error t value
1|2 -1.2246  0.4425    -2.7675
2|3  1.0072  0.3884     2.5931
3|4  2.8052  0.5101     5.4987
4|5  4.0133  0.6135     6.5411

Residual Deviance: 189.7068 
AIC: 201.7068 
Next step: Use 'setx' method

> summary(z.out2)
Model: 
Call:
z5$zelig(formula = rating ~ contact + temp, data = w.amelia$imputations$imp2)

Coefficients:
           Value Std. Error t value
contactyes  1.73     0.4760   3.635
tempwarm    1.69     0.4774   3.539

Intercepts:
    Value   Std. Error t value
1|2 -0.6469  0.4850    -1.3338
2|3  1.3290  0.4659     2.8525
3|4  2.8718  0.5571     5.1547
4|5  4.2483  0.6751     6.2932

Residual Deviance: 198.7817 
AIC: 210.7817 
Next step: Use 'setx' method

R: Ordered logistic regression with multiple imputation data (amelia package)

1 Answers