Variable order requirement for link, sim, ensemble

[rmcelreath]

Variable order requirement for link, sim, ensemble

[rmcelreath]

It's not that they require the order to be the same. It's that the data frame will not be defined with the correct number of rows, unless the first variable you define has the right length. Putting sd.growing.season, in your example, first define the correct length. Then R will fill out the single values to all rows.

You could also just make sure each variable is the right length:

dpred = data.frame(mean.growing.season = rep(11,100),
                   sd.growing.season = seq(0, 6, len = 100),
                   logArea = rep(mean(n$logArea),100))

[michaellevy]

I think data.frame uses the longest supplied vector to determine the number of rows. Here it does get it right:

> dpred = data.frame(mean.growing.season = 11,
+                    sd.growing.season = seq(0, 6, len = 100),
+                    logArea = mean(n$logArea))
> nrow(dpred)
[1] 100

Also, explicitly defining the length of each column doesn't fix the issue:

Wrong order, implied length:

> m = lm(loglpc ~ mean.growing.season * sd.growing.season + log10(area), data = n)
> dpred = data.frame(mean.growing.season = 11,
+                    sd.growing.season = seq(0, 6, len = 100),
+                    logArea = mean(n$logArea))
> l = link(m, dpred)
> summary(apply(l, 2, mean))
   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
 -3.615  -3.112  -2.608  -2.608  -2.105  -1.601 

Right order, implied length:

> dpred = dpred[, 3:1]
> l = link(m, dpred)
> summary(apply(l, 2, mean))
   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
 -2.753  -2.628  -2.502  -2.502  -2.377  -2.251 

Wrong order, explicit length:

> m = lm(loglpc ~ mean.growing.season * sd.growing.season + log10(area), data = n)
> dpred = data.frame(mean.growing.season = rep(11, 100),
+                    sd.growing.season = seq(0, 6, len = 100),
+                    logArea = rep(mean(n$logArea), 100))
> l = link(m, dpred)
> summary(apply(l, 2, mean))
   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
 -3.640  -3.126  -2.612  -2.612  -2.098  -1.584 

[rmcelreath]

I misunderstood the original issue. Is this behavior that happens only with lm?

The link method for lm is not something I ever finished writing. I could didn't get it to behave right, and couldn't justify spending time on it.

[michaellevy]

Yes, the issue seems to be constrained to link for lm's. Maybe add at least a warning to link when called with an lm?

[rmcelreath]

I should either fix it or disable it. Might be an easy fix. But bug seems to arise from the way I've hacked into the predict method. So will take a little detective work.