0
$\begingroup$

I have a model with Machine as factorial variable. The contrast is set to "contr.sum".

library(MEMSS)
#> Warning: package 'MEMSS' was built under R version 3.6.3
#> Loading required package: lme4
#> Loading required package: Matrix
#> 
#> Attaching package: 'MEMSS'
#> The following objects are masked from 'package:datasets':
#> 
#>     CO2, Orange, Theoph
str(Machines)
#> 'data.frame':    54 obs. of  3 variables:
#>  $ Worker : Factor w/ 6 levels "1","2","3","4",..: 1 1 1 2 2 2 3 3 3 4 ...
#>  $ Machine: Factor w/ 3 levels "A","B","C": 1 1 1 1 1 1 1 1 1 1 ...
#>  $ score  : num  52 52.8 53.1 51.8 52.8 53.1 60 60.2 58.4 51.1 ...
fm <- lm(score ~ Machine, Machines, contrasts = list(Machine = "contr.sum"))
summary(fm)
#> 
#> Call:
#> lm(formula = score ~ Machine, data = Machines, contrasts = list(Machine = "contr.sum"))
#> 
#> Residuals:
#>      Min       1Q   Median       3Q      Max 
#> -17.3222  -2.1431   0.4444   4.4403   9.3778 
#> 
#> Coefficients:
#>             Estimate Std. Error t value Pr(>|t|)    
#> (Intercept)  59.6500     0.7861  75.884  < 2e-16 ***
#> Machine1     -7.2944     1.1117  -6.562 2.68e-08 ***
#> Machine2      0.6722     1.1117   0.605    0.548    
#> ---
#> Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
#> 
#> Residual standard error: 5.776 on 51 degrees of freedom
#> Multiple R-squared:  0.5077, Adjusted R-squared:  0.4884 
#> F-statistic:  26.3 on 2 and 51 DF,  p-value: 1.415e-08

How can I use the predict function to calculate the average over all machines? Just like I can do it for machine A:

pred_dat <- data.frame(Machine = "A")
predict(fm, newdata = pred_dat)
#>        1 
#> 52.35556

This, of course, will not work. But it would be nice to have it that simple.

pred_dat <- data.frame(Machine = "mean_over_all_machines")
predict(fm, newdata = pred_dat)
#> Error in model.frame.default(Terms, newdata, na.action = na.action, xlev = object$xlevels): factor Machine has new level mean_over_all_machines

The most obious way to get the mean over all levels is to just take the intercept, which is the mean. But I want to use the predict function with complex mixed models, many variables, interactions and random effects etc. (in this case glmmTMB). So, writing the complete term for the calculation of the average is not really an option.

For me this seems like a very general functionality and I am wondering why there is no option to predict without considering the categorical level.

Improve this question
$\endgroup$
5
  • $\begingroup$ pred_dat <- data.frame(Machine = levels(Machines$machine)) ; mean(red_dat) ; i don't know if you are referring to the mean of the predicted values for all machines $\endgroup$
    – StupidWolf
    Commented Apr 8, 2020 at 12:12
  • $\begingroup$ I also considered averaging the predictions afterwards. But I am uncertain if that may lead to false predictions under some cirumstances. If not, this of course solves the problem. $\endgroup$
    – Hans Jürgen
    Commented Apr 8, 2020 at 12:15
  • $\begingroup$ what is a false prediction? anyway, you should note that in this case, since you have only 1 categorical variable, the intercept is the mean for machine0 (the missing level) $\endgroup$
    – StupidWolf
    Commented Apr 8, 2020 at 12:18
  • $\begingroup$ you can check with tapply(Machines$score , Machines$machine ,mean) $\endgroup$
    – StupidWolf
    Commented Apr 8, 2020 at 12:18
  • $\begingroup$ Sorry I meant wrong prediction. Ok, I will assume averaging afterwards is not problematic. So far I am not able to make up an example which leads to wrong prediction when averaging afterwards. Still it would be usefull to do this with the predict function i.e. for calculating standard errors. $\endgroup$
    – Hans Jürgen
    Commented Apr 8, 2020 at 12:40

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.