No difference in R package MuMIn model.avg output when using delta < 2 vs. 95% AICc weight?

Question

I've developed a set of candidate models (several binomial GLMs with a null and two global models) and I am trying to use MuMIn to do model averaging on the top delta 2 AIC models.

Here is a subset of the candidate models

fit1  <- glm(formula = Y ~ lnRd + offset(log(nights_active)), binomial, df)
fit2  <- glm(formula = Y ~ sqRC + offset(log(nights_active)), binomial, df) 
fit3  <- glm(formula = Y ~ DT_River + offset(log(nights_active)), binomial, df)
fit4  <- glm(formula = Y ~ lnMB + offset(log(nights_active)), binomial, df)
fit5  <- glm(formula = Y ~ cbFI + offset(log(nights_active)), binomial, df)
fit6  <- glm(formula = Y ~ DT_CM + offset(log(nights_active)), binomial, df)
fit7  <- glm(formula = Y ~ sqVl + offset(log(nights_active)), binomial, df)
fit8  <- glm(formula = Y ~ DT_Human + offset(log(nights_active)), binomial, df)
fit9 <- glm(formula = Y ~ lnPv + offset(log(nights_active)), binomial, df)
...
fitNull <- glm(formula = Y ~ 1 + offset(log(nights_active)), binomial, df)
fitGb1  <- glm(formula = Y ~ DT_River + lnMB + DT_Human + lnRd + sqVl + offset(log(nights_active)), binomial, df)
fitGb2  <- glm(formula = Y ~ sqRC + DT_CM + cbFI + lnPv + offset(log(nights_active)), binomial, df)

cand.set<-list(fit1,fit2,fit3,fit4,fit5,fit6,fit7,fit8,fit9,...fitNull,fitGb1,fitGb2)

And here is the code I am using:

avg.mod <- model.avg(cand.set, subset = delta < 2, fit = TRUE, revised.var = TRUE)

summary(avg.mod)
confint(avg.mod)

But when I look at the summary results it looks like all of my covariates are included, not just ones in the top 2 delta AIC? I confirmed this by comparing results when I changed the subset to top delta 4 and top 95% of AIC wi with cumsum(weight) <= .95. No matter what, the summary shows the same results - with all of my covariates.

Call:
model.avg(object = cand.set, subset = delta < 2, fit = FALSE, 
    revised.var = FALSE)

Component model call: 
glm(formula = <25 unique values>, family = binomial, data = df)

Component models: 
             df logLik  AICc delta weight
2/4/8         3 -16.92 40.31  0.00   0.18
4/8           2 -18.13 40.49  0.18   0.16
3/8           2 -18.31 40.85  0.55   0.14
1/4/8         3 -17.74 41.96  1.65   0.08
4/5/8         3 -17.81 42.10  1.79   0.07
4/7/8         3 -18.13 42.72  2.42   0.05
4/8/10        3 -18.13 42.73  2.42   0.05
2/8           2 -19.28 42.79  2.49   0.05
8             1 -20.68 43.43  3.12   0.04
5/8           2 -19.85 43.94  3.63   0.03
6/8           2 -20.21 44.65  4.35   0.02
8/9           2 -20.44 45.11  4.80   0.02
7/8           2 -20.52 45.26  4.96   0.01
8/10          2 -20.66 45.55  5.25   0.01
1/8           2 -20.67 45.58  5.27   0.01
5/8/9         3 -19.60 45.68  5.37   0.01
5/7/8         3 -19.79 46.05  5.75   0.01
5/8/10        3 -19.85 46.18  5.87   0.01
6/8/9         3 -20.09 46.64  6.34   0.01
7/8/9         3 -20.20 46.88  6.57   0.01
6/8/10        3 -20.21 46.89  6.59   0.01
1/8/9         3 -20.43 47.33  7.03   0.01
1/2/6/8/9     5 -18.10 47.42  7.11   0.01
7/8/10        3 -20.50 47.47  7.16   0.00
3/4/5/7/8/10  6 -17.36 48.47  8.16   0.00

Term codes: 
                      cbFI                      DT_CM                   DT_Human                   DT_River 
                         1                          2                          3                          4 
                      lnMB                       lnPv                       lnRd offset(log(nights_active)) 
                         5                          6                          7                          8 
                      sqRC                       sqVl 
                         9                         10 

Model-averaged coefficients:  
(full average) 
              Estimate Std. Error Adjusted SE z value Pr(>|z|)    
(Intercept) -4.607e+00  3.892e-01   3.983e-01  11.567   <2e-16 ***
DT_River    -5.231e-01  5.378e-01   5.425e-01   0.964    0.335    
DT_CM        1.416e-01  2.499e-01   2.514e-01   0.563    0.573    
DT_Human     1.144e-01  2.125e-01   2.132e-01   0.536    0.592    
cbFI        -2.933e-02  8.389e-02   8.478e-02   0.346    0.729    
lnMB        -5.850e-02  1.283e-01   1.293e-01   0.452    0.651    
lnRd         9.263e-03  4.494e-02   4.573e-02   0.203    0.839    
sqVl        -5.795e-06  3.338e-02   3.417e-02   0.000    1.000    
lnPv         1.443e-02  3.540e-02   3.570e-02   0.404    0.686    
sqRC        -1.369e-02  3.695e-02   3.735e-02   0.366    0.714    
 
(conditional average) 
              Estimate Std. Error Adjusted SE z value Pr(>|z|)    
(Intercept) -4.607e+00  3.892e-01   3.983e-01  11.567   <2e-16 ***
DT_River    -8.697e-01  4.263e-01   4.362e-01   1.994   0.0462 *  
DT_CM        6.027e-01  3.904e-01   3.996e-01   1.508   0.1315    
DT_Human     8.245e-01  4.137e-01   4.232e-01   1.948   0.0514 .  
cbFI        -2.894e-01  5.025e-01   5.125e-01   0.565   0.5722    
lnMB        -4.287e-01  4.337e-01   4.436e-01   0.966   0.3338    
lnRd         9.963e-02  3.830e-01   3.917e-01   0.254   0.7992    
sqVl        -6.426e-05  3.701e-01   3.788e-01   0.000   0.9999    
lnPv         3.652e-01  4.167e-01   4.265e-01   0.856   0.3919    
sqRC        -2.589e-01  3.821e-01   3.911e-01   0.662   0.5079    
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Is there something I am missing here? It even shows in the summary that the delta values span over 8, not 2...

Answer 1

I figured this out! Apparently R doesn't like the candidate set as type "list". I ran again with this code, which saves the candidate set using "model.sel" instead, and now it works.

cand.set<-model.sel(fit1,fit2,fit3,fit4,fit5,fit6,fit7,fit8,fit9,fit10,fit11,fit12,fit13,fit14,fit15,fit16,fit17,fit18,fit19,fit20,fit21,fit22,fitNull,fitGb1,fitGb2)

selall<-model.sel(cand.set)
avgall <- model.avg(selall, subset = delta < 2)
summary(avgall)