0 proportion showing up as NA in props() with df_stats()

[rpruim]

require(Lock5withR)
df_stats( Empathy ~ Sex, data = CompassionateRats, props)
##   Sex  prop_yes   prop_no
## 1   F 1.0000000        NA
## 2   M 0.7083333 0.2916667

[dtkaplan]

@rpruim I took care of this to some extent when I merged issue008 with beta. I didn't see an easy way to detect when counts() is being used in df_stats() and then to fix it there. And I don't think we can rely on NA always turning into 0 for general statistics. So I modified counts() to change NA to 0. This percolates through to props(), which is based on counts().

In df_stats(), the flag long_names is TRUE by default. so the names I am getting with the beta branch are

props(Empathy ~ Sex, data = CompassionateRats)
  Sex Empathy_prop_yes Empathy_prop_no
1   F        1.0000000       0.0000000
2   M        0.7083333       0.2916667

Of course you can change this, but I'm not sure why your names by default correspond to long_names = FALSE

> props(Empathy ~ Sex, data = CompassionateRats, long_names = FALSE)
  Sex  prop_yes   prop_no
1   F 1.0000000 0.0000000
2   M 0.7083333 0.2916667

[rpruim]

Thanks. I've been busy with other things and haven't had a chance to take a closer look at issue008.

[rpruim]

After merging stuff, I'm getting

props(Empathy ~ Sex, data = CompassionateRats)
##   Sex Empathy_prop_yes Empathy_prop_no
## 1   F        1.0000000       0.0000000
## 2   M        0.7083333       0.2916667
require(Lock5withR)
df_stats( Empathy ~ Sex, data = CompassionateRats, props)
##   Sex props_Empathy_prop_yes props_Empathy_prop_no
## 1   F              1.0000000                    NA
## 2   M              0.7083333             0.2916667

Note: the fix isn't working inside a call to df_stats() and the naming of the columns has changed since what I had before the merge.

Perhaps I misunderstood that this was ready to go and waiting for me to merge with master.

In any case, this issue is definitely still open.

[rpruim]

I've reset master back to before this whole process began. I think we should start from scratch to fix this since (a) I messed things up in the repo and (b) the proposed fix was not a complete solution anyway. Certainly we can look back at what was done for ideas/code as needed.

[rpruim]

I have a partial solution -- it works for factors (since it can use the levels to know to compute the 0's). So the original example works because Empathy is a factor with levels yes and no.

options(digits = 3)
require(Lock5withR)
df_stats( Empathy ~ Sex, data = CompassionateRats, props)
##   Sex prop_no prop_yes
## 1   F   0.000    1.000
## 2   M   0.292    0.708

The current implementation of df_stats() computes group by group without knowing what is going on in the other groups -- including the possible values of the summarized variable. The NA in the table gets introduced at a later stage, when the results are brought into a table. As @dtkaplan suggests, just converting NAs to 0's generally probably isn't the best solution.