Does it matter to "goodness of the model" what values are used as references in factor data?

Asked Oct 14 '16 at 11:25

Active Oct 14 '16 at 11:25

Viewed 33 times

The reason for asking is that when using factor data such as in the following post:
How to interpret Pr(>|t|) of factor variables?

one can adjust the p-values of independent variables of the factor data by changing the reference value, e.g. by using R's relevel() function. But even if this does change the p-values to make variables seem more significant, then does changing the reference actually change the model or the goodness of the model in any way?

edited Apr 13 '17 at 12:44

Community

asked Oct 14 '16 at 11:25

mavavilj

3,167
4
20
44

Indifference to an arbitrary choice of reference level is clearly desirable, but it depends rather how you're fitting the model. – Scortchi - Reinstate Monica Oct 14 '16 at 13:44
@Scortchi Can you eloaborate? – mavavilj Oct 14 '16 at 13:49
If you're penalizing coefficients according to their magnitude, for example. I meant to suggest *you* elaborate on the model & fitting procedure you're considering. – Scortchi - Reinstate Monica Oct 14 '16 at 13:53
@Scortchi So what info do you need? The `lm()`command to create the model and a description of the response and predictors? – mavavilj Oct 14 '16 at 14:02
"A linear model fit by ordinary least squares". (Though a good answer might well explain when the coding makes a substantive difference & when it doesn't.) – Scortchi - Reinstate Monica Oct 14 '16 at 14:18

Does it matter to "goodness of the model" what values are used as references in factor data?

0 Answers0