pander.lm() and pander.summary.lm(): Reordering of coefficients

[krlmlr]

The coefficient that appears first in the output of summary.lm() seems to appear last in the output of pander::pander.lm(). It looks like this is by design, but this makes models without intercept look very awkward.

> lm(Petal.Width~Petal.Length+Sepal.Width+Species-1, iris) %>% pander::pander

--------------------------------------------------------------------
                    Estimate   Std. Error   t value   Pr(>|t|) 
----------------------- ---------- ------------ --------- ----------
    **Sepal.Width**       0.202       0.0442      4.571    1.03e-05 

   **Speciessetosa**     -0.6953      0.1424     -4.882   2.739e-06 

 **Speciesversicolor**   0.04165      0.1546     0.2695     0.7879  

 **Speciesvirginica**     0.4806      0.1901      2.529    0.01251  

   **Petal.Length**       0.1701     0.03489      4.876   2.808e-06 
--------------------------------------------------------------------

Table: Fitting linear model: Petal.Width ~ Petal.Length + Sepal.Width + Species - 1

> lm(Petal.Width~Petal.Length+Sepal.Width+Species-1, iris) %>% summary

Call:
lm(formula = Petal.Width ~ Petal.Length + Sepal.Width + Species - 
    1, data = iris)

Residuals:
     Min       1Q   Median       3Q      Max 
-0.55861 -0.08486 -0.00190  0.09073  0.48605 

Coefficients:
                  Estimate Std. Error t value Pr(>|t|)    
Petal.Length       0.17012    0.03489   4.876 2.81e-06 ***
Sepal.Width        0.20203    0.04420   4.571 1.03e-05 ***
Speciessetosa     -0.69529    0.14242  -4.882 2.74e-06 ***
Speciesversicolor  0.04165    0.15457   0.269   0.7879    
Speciesvirginica   0.48064    0.19005   2.529   0.0125 *  

---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 0.1685 on 145 degrees of freedom
Multiple R-squared:  0.9864,    Adjusted R-squared:  0.9859 
F-statistic:  2100 on 5 and 145 DF,  p-value: < 2.2e-16

[daroczig]

Good catch, thanks for reporting this issue! The above (very trivial, but hopefully OK) commit should resolve this behavior -- looking forward to any feedback.

[krlmlr]

Thanks. I'd prefer not to rearrange parameters at all (or at least make it optional), but you probably have a good reason to do this the way you're doing it.

[daroczig]

@RomanTsegelskyi do you remember the motives behind moving the intercept to the bottom of the table?

[romantseg]

I can't recall the reason right away, so it might be my mistake when I was refactoring the implementation

[daroczig]

I spent some time digging the git history and found d28cf55 to be the related commit. This is due to #46, and we are moving the intercept to the bottom of the table as this looked appropriate based on the stargazer behavior.

@krlmlr @kovla does it make sense? Not sure if it's needed or not (personally, I do not care if the intercept is on the top or at the bottom of the table), looking forward to any feedback. And of course we can make this optional.

[krlmlr]

IMO, if some journals require this behavior, it's reasonable to support it as an option. Making it default confuses newbies that are used to R's output.

[daroczig]

Makes sense, done with the above commit:

> pander(lm(am ~ wt, mtcars))

--------------------------------------------------------------
              Estimate   Std. Error   t value   Pr(>|t|) 
----------------- ---------- ------------ --------- ----------
 **(Intercept)**    1.542       0.2256      6.838   1.378e-07 

     **wt**        -0.3532     0.06717     -5.258   1.125e-05 
--------------------------------------------------------------

Table: Fitting linear model: am ~ wt

> pander(lm(am ~ wt, mtcars), move.intercept = TRUE)

--------------------------------------------------------------
              Estimate   Std. Error   t value   Pr(>|t|) 
----------------- ---------- ------------ --------- ----------
     **wt**        -0.3532     0.06717     -5.258   1.125e-05 

 **(Intercept)**    1.542       0.2256      6.838   1.378e-07 
--------------------------------------------------------------

Table: Fitting linear model: am ~ wt