如何通过 TOBIT model（censReg 包）的对比进行事后分析？

Question

This question was migrated from Stack Overflow because it can be answered on Cross Validated. Migrated 7 days ago .

I have 4 groups of animals that did a task on four consecutive days. The variable of interest was latency, which was censored at 60 seconds. I fit a mixed effects linear censored regression model as follows:

library(censReg)
m <- censReg(Latency ~ Group * Day,
  data = data, method = "BHHH", right = 60)
summary(m)

Observations:
         Total  Left-censored     Uncensored Right-censored 
           608              0            482            126 

Coefficients:
             Estimate Std. error t value  Pr(> t)    
(Intercept)  59.64948    4.80290  12.419  < 2e-16 ***
Group2        1.82896    6.22570   0.294  0.76893    
Group3      -26.53572    6.32002  -4.199 2.68e-05 ***
Group4       -4.24361    6.61237  -0.642  0.52102    
Day         -11.98992    1.39992  -8.565  < 2e-16 ***
Group2:Day   -0.63445    1.66085  -0.382  0.70246    
Group3:Day    6.56444    2.05358   3.197  0.00139 ** 
Group4:Day    3.52366    2.42819   1.451  0.14674    
logSigmaMu    2.13480    0.19363  11.025  < 2e-16 ***
logSigmaNu    3.04728    0.03928  77.577  < 2e-16 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

BHHH maximisation, 53 iterations
Return code 8: successive function values within relative tolerance limit (reltol)
Log-likelihood: -2302.083 on 10 Df



library(parameters)
model_parameters(m)
         Parameter Coefficient         SE   CI     CI_low    CI_high          t df_error             p
    1  (Intercept)  59.6494822 4.80289885 0.95  50.216882  69.082082 12.4194750      598  1.140841e-31
    2       Group2   1.8289648 6.22570236 0.95 -10.397934  14.055864  0.2937765      598  7.690307e-01
    3       Group3 -26.5357235 6.32002420 0.95 -38.947865 -14.123582 -4.1986743      598  3.093431e-05
    4       Group4  -4.2436141 6.61237005 0.95 -17.229905   8.742677 -0.6417690      598  5.212691e-01
    5          Day -11.9899227 1.39992101 0.95 -14.739282  -9.240563 -8.5647137      598  9.194472e-17
    6   Group2:Day  -0.6344538 1.66084945 0.95  -3.896261   2.627353 -0.3820056      598  7.025929e-01
    7   Group3:Day   6.5644385 2.05358314 0.95   2.531327  10.597550  3.1965779      598  1.464037e-03
    8   Group4:Day   3.5236628 2.42818993 0.95  -1.245154   8.292479  1.4511479      598  1.472630e-01
    9   logSigmaMu   2.1347957 0.19362696 0.95   1.754524   2.515067 11.0253019      598  7.364901e-26
    10  logSigmaNu   3.0472752 0.03928047 0.95   2.970131   3.124420 77.5773541      598 2.539111e-314

I don't know how to do a post-hoc analysis to compare all groups with one another on each day. I always use the emmeans package to get contrasts, but it doesn't work for a censReg object. Do you have any suggestions to get around that? Easy solutions are most welcome:-) You can get the data in a csv file here .

---

As suggested, I tried with the qdrg() funciton from the emmeans package.

rg <- qdrg(Latency ~ Group * Day, data=data, coef=coef(m),vcov=vcov(m), df=m$df.residual)

'emmGrid' object with variables:
        Group = 1, 2, 3, 4
        Day = 2.5

Although the qdrg() funciton seemed to work, the rg grid did not work with the emmenas function.

emmeans(rg, pairwise ~ Group|Day,  at = list(Day = c(1,2,3, 4)), adjust="none")

Error in X[ii, ii, drop = FALSE] %*% y[ii] : non-conformable arguments

I don't know how to fix it.

Answer 1

As suggested, I tried with the qdrg() funciton from the emmeans package.

rg <- qdrg(Latency ~ Group * Day, data=data, coef=coef(m),vcov=vcov(m), df=m$df.residual)

'emmGrid' object with variables:
        Group = 1, 2, 3, 4
        Day = 2.5

Although the qdrg() funciton seemed to work, the rg grid did not work with the emmenas function.

emmeans(rg, pairwise ~ Group|Day,  at = list(Day = c(1,2,3, 4)), adjust="none")

Error in X[ii, ii, drop = FALSE] %*% y[ii] : non-conformable arguments

I don't know how to fix it.

Answer 2

To make qdrg() work, I needed to subset the coefficients.

coef <- coef(m)[1:8]
vcov <- vcov(m)[1:8, 1:8]
rg <- qdrg(Latency ~ Group * Day, data=data, coef=coef, vcov=vcov, df=m$df.residual, at = list(Day = c(1,2,3,4)))
emmeans(rg, pairwise ~ Group|Day, adjust="none")