Including number of observations used to build the model about broom

[alexpghayes]

Closed in #597! Thanks @vincentarelbundock!

[grasshoppermouse]

I also think it would be useful if glance output included nobs. It looks like glance.stanreg already does:

https://github.com/dgrtwo/broom/blob/master/R/rstanarm_tidiers.R

[hughjonesd]

Just adding my vote to this feature. I think in most fields, it is standard to report the N as one of the summary statistics.

[randomgambit]

hello there! is there a fix for that very important feature?
thanks!!

[alexpghayes]

I'd be willing to this in a tryCatch to finish_glance to pick up an n column for relevant models. The question then is how many models actually implement nobs() methods.

I'm hesitant to report counts for each factor level in glance() because this is moving more into data summarizing than properties of a model. If a tidy() method doesn't already inherit this information from summary() or whatnot I don't think it's worth the effort to try implement this consistently across tidiers. Also, skimr::skim() is fantastic for this sort of thing.

[vincentarelbundock]

I'm building a regression summary package built on broom and users are requesting this feature. It's pretty important to me, and I would be willing to do it if you tell me about your implementation preferences.

To answer your question, a lot of models actually implement the nobs method. I went through every extract method in the texreg package. I may have missed a couple, but the models not listed below should work with nobs:

#default
stats::nobs(model)

#felm
summary(model)$N

#censReg
summary(model)$nobs

#btergm
#mbtergm
model@nobs

#betaor
#betamfx
model$fit$nobs

#averaging
#model.selection (MuMIn)
as.numeric(attr(model, 'nobs'))

#sienaFit
model$n

#zeroinfl
summary(model)$n

#fGARCH
length(model@data) 

# gel
NROW(model$gt) 

# lme4
dim(model.frame(model))[1] 

#lmrob
#systemfit
#lmRob
length(model$residuals) 

#logitmfx
#probitmfx
#negbinirr
#negbinmfx
nrow(model$fit$model) 

#lrm
model$stats[1]

#mlogit
#plm
#pmg
#rq
#summary.lm
nrow(summary(model)$residuals)

#mnlogit
s$model.size$N

#multinom
#sarlm
nrow(summary(model$fitted.values))

#pgmm
attr(summary(model), 'pdim')$nT$N

#simex
length(model$model$residuals)

#survreg
length(model$linear.predictors)

#zelig
nrow(model$data)

#pglm
length(model$gradientObs[, 1])

[vincentarelbundock]

@alexpghayes One possible design that I would be willing to implement:

For each element of the list above which are not compatible with nobs, extract the value explicitly in the model-specific glance function.

Modify the finish_glance function. If 'n' %in% names(ret), then tryCatch(stats::nobs)

This is a bit more work (which I am willing to do), but it's explicit, and would avoid unexpected side effects from having a bunch of ifelse statements.

[gavinsimpson]

Sounds simpler to just implement nobs() methods for these internally, where they don't exist. Ideally these would be offered upstream to the respective package maintainers, but there's nothing stopping these being in broom if they are not wanted or maintainers are unresponsive. Using nobs() makes it simpler/safer to implement return of number of observations in glance.

[vincentarelbundock]

Sure. I'm happy to write a bunch of nobs methods if people find those useful. There's also a discussion about how these methods would be used in my WIP PR: #594

Basically,

1. Each glance function calls its own nobs()
2. finish_glance calls nobs() for all of them, as it currently does for AIC, BIC, etc.

[vincentarelbundock]

I'm in the process of checking every model object for which broom offers a glance function to see if they work with stats::nobs. I'm also writing new methods for those that don't. The results are collected in this Gist:

https://gist.github.com/vincentarelbundock/24bedac98499181790aab230cc5b74bc

[github-actions]

This issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with a reprex: https://reprex.tidyverse.org) and link to this issue.