1

I'm trying to extract two different averaged models from MuMIn for output to latex via texreg or stargazer. I'd like to have one table where I can compare two species' response to different sets of abiotic variables, that looks the same as one created from two model objects using

glmtable <- texreg(list(m1, m2).

The above code will work on glm objects but not on averaged model objects created in MuMIn.

I tried following an example at https://sites.google.com/site/rforfishandwildlifegrads/home/mumin_usage_examples, to output a text table that can be output to latex.

Here's a reproducible example using the cement data:

library(MuMIn)
data(cement)

# full model
fm1 <- lm(y ~ ., data = Cement, na.action = na.fail)
# create and examine candidate models
(ms1 <- dredge(fm1))

# average models with delta AICc <5, include adjusted SE
MA.ests<-model.avg(ms1, subset= delta < 5, revised.var = TRUE)

This works fine. However when I call 

MA.ests$avg.model

I get >NULL.

Has avg.model been deprecated? Or is there another way to do this?

I can do a workaround using any of these three calls, but they're not exactly what I want.

coefTable(MA.ests)
coef(MA.ests)
modavg.table <- as.data.frame(summary(MA.ests)$coefmat)

(that is, I don't know how to get these objects into latex without a lot more code.)

Thanks in advance for any suggestions.

www
  • 38,575
  • 12
  • 48
  • 84
cleo wolfki
  • 13
  • 6
  • `ms1` is a `"model.selection"` object, and you meant to use `MA.ests` the `"averaging"` object returned by `model.avg`. Also, it is generally bad idea to access the objects elements directly (the internal object structure is subject to change), use extractor functions like `coefTable` and `confint` instead. – Kamil Bartoń Mar 09 '15 at 22:56
  • Thanks, @Kamil, I edited the post--however still get NULL when I run the code. Is there a way to extract a model object that can be summarized using any of the latex tools? – cleo wolfki Mar 10 '15 at 01:47
  • The former `"avg.model"` element has been renamed to `"coefTable"` (now I notice that this is not reflected in the manual). Safer way of getting the summary table may be e.g. `cbind(coefTable(x), confint(x))`. – Kamil Bartoń Mar 10 '15 at 09:26
  • It should be straightforward to write an extension for texreg for ``model.selection`` objects. You may want to read section 6 of the texreg article in the Journal of Statistical Software (updated version: http://cran.r-project.org/web/packages/texreg/vignettes/v55i08.pdf) for details on how to do this. – Philip Leifeld Mar 10 '15 at 12:27
  • Ok! This looks promising. Thanks for the texreg link as well...I am still cutting my teeth on r and got a bit intimidated by section 6, but will persevere and post the code here when I've got it. Thanks again. – cleo wolfki Mar 10 '15 at 18:11

2 Answers2

1

The latest version 1.34.3 of the texreg package supports both model.selection and averaging objects.

Your code example:

library("texreg")
library("MuMIn")
data(cement)
fm1 <- lm(y ~ ., data = Cement, na.action = na.fail)
ms1 <- dredge(fm1)

screenreg(ms1)

yields:

==========================================================================================================================================================================================================
                Model 1     Model 2     Model 3     Model 4     Model 5     Model 6     Model 7   Model 8     Model 9     Model 10    Model 11    Model 12  Model 13    Model 14    Model 15    Model 16  
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
(Intercept)      52.58 ***   71.65 ***   48.19 ***  103.10 ***  111.68 ***  203.64 ***   62.41    131.28 ***   72.07 ***  117.57 ***   57.42 ***   94.16     81.48 ***   72.35 ***  110.20 ***   95.42 ***
                 (2.29)     (14.14)      (3.91)      (2.12)      (4.56)     (20.65)     (70.07)    (3.27)      (7.38)      (5.26)      (8.49)     (56.63)    (4.93)     (17.05)      (7.95)      (4.17)   
X1                1.47 ***    1.45 ***    1.70 ***    1.44 ***    1.05 ***                1.55 *                                                              1.87 ***    2.31 *                          
                 (0.12)      (0.12)      (0.20)      (0.14)      (0.22)                  (0.74)                                                              (0.53)      (0.96)                           
X2                0.66 ***    0.42 *      0.66 ***                           -0.92 ***    0.51                  0.73 ***                0.79 ***    0.31                                                  
                 (0.05)      (0.19)      (0.04)                              (0.26)      (0.72)                (0.12)                  (0.17)      (0.75)                                                 
X4                           -0.24                   -0.61 ***   -0.64 ***   -1.56 ***   -0.14     -0.72 ***               -0.74 ***               -0.46                                                  
                             (0.17)                  (0.05)      (0.04)      (0.24)      (0.71)    (0.07)                  (0.15)                  (0.70)                                                 
X3                                        0.25                   -0.41 *     -1.45 ***    0.10     -1.20 ***   -1.01 ***                                                  0.49       -1.26 *              
                                         (0.18)                  (0.20)      (0.15)      (0.75)    (0.19)      (0.29)                                                    (0.88)      (0.60)               
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Log Likelihood  -28.16      -26.93      -26.95      -29.82      -27.31      -29.73      -26.92    -35.37      -40.96      -45.87      -46.04      -45.76    -48.21      -48.00      -50.98      -53.17    
AICc             69.31       72.44       72.48       72.63       73.19       78.04       79.84     83.74       94.93      100.41      100.74      104.52    105.08      109.01      110.63      111.54    
Delta             0.00        3.13        3.16        3.32        3.88        8.73       10.52     14.43       25.62       31.10       31.42       35.21     35.77       39.70       41.31       42.22    
Weight            0.57        0.12        0.12        0.11        0.08        0.01        0.00      0.00        0.00        0.00        0.00        0.00      0.00        0.00        0.00        0.00    
Num. obs.        13          13          13          13          13          13          13        13          13          13          13          13        13          13          13          13       
==========================================================================================================================================================================================================
*** p < 0.001, ** p < 0.01, * p < 0.05

And model averaging:

MA.ests <- model.avg(ms1, subset = delta < 5, revised.var = TRUE)

screenreg(MA.ests)

yields:

=======================
             Model 1   
-----------------------
(Intercept)   64.69 ** 
             (22.24)   
X1             1.46 ***
              (0.20)   
X2             0.63 ***
              (0.12)   
X4            -0.48 *  
              (0.22)   
X3            -0.02    
              (0.38)   
-----------------------
Num. obs.     13       
=======================
*** p < 0.001, ** p < 0.01, * p < 0.05

For finetuning, see also the arguments of the two extract methods on the help page: ?extract

Philip Leifeld
  • 2,253
  • 15
  • 24
  • Just a note - this only seems to extract the 'full' average, not the conditional average – bjw Oct 21 '16 at 11:58
0

If you don't want the dependency you can get the p values directly from the summary object:

summary(averagingobject)$coefmat.full

summary(averagingobject)$coefmat.subset
bjw
  • 2,046
  • 18
  • 33