Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, is already up and running.

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: How to compare RMSE of two models

From   Nick Cox <>
Subject   Re: st: How to compare RMSE of two models
Date   Tue, 27 Nov 2012 13:14:53 +0000

That seems to me the same question re-worded, but not more clearly. As
often happens, I started writing a reply to your first post broadly
similar to Maarten's, but he got there first.

You can use RMSE as a criterion, in which case the model with the
lower RMSE is the better model. Naturally many would want to emphasise
here that e.g. a scatter plot with fitted regression line would be as
or more informative about which model is "better".

Recasting that comparison as a signficance test seems to raise more
questions than it solves, not least quite what sense can be given to a
significance test given bootstrap results for RMSE for two models with
a shared response and different predictors.

If people don't understand what you want and/or think it seems unsound
in principle, they won't usually pass to the next question of
suggesting Stata commands to do it.


On Tue, Nov 27, 2012 at 12:49 PM, rasool.bux <> wrote:
> Dear Maarten,
> Thanks for reply. May be I am not conveying my question clearly. I would like to compare the two models against the same yvariable, which model is performing better and how we could say that model 1 is better than model 2 i.e. in terms of p-values of RMSE.
> Best regards
> Rasool Bux
> -----Original Message-----
> From: [] On Behalf Of Maarten Buis
> Sent: Tuesday, November 27, 2012 2:45 PM
> To:
> Subject: Re: st: How to compare RMSE of two models
> On Tue, Nov 27, 2012 at 10:32 AM, rasool.bux wrote:
>> How to statistical compare RMSE of two equations/models in Stata? Kindly let me know which test used for it and how to report it. Same for IQR (95%CIs) reporting as single value i.e. p75 - p25.
>> bootstrap "regress yvar xvar1" _b rmse1=e(rmse), reps(1000)
>> saving(rmse1) bootstrap "regress yvar xvar2" _b rmse2=e(rmse),
>> reps(1000) saving(rmse2)
> The first thing to consider when you want to perform a statistical test is to think about the null hypothesis. In this case I don't think what you want to do involves a meaningful null hypothesis. In that case you just don't want to perform that test.
>> bootstrap iqr_v1=(r(p75)-r(p25)), rep(1000) seed(12345789) nodots
>> saving(iqr_v1,replace): summarize v1, detail
> Again consider the null hypothesis, and think whether you want to test that hypothesis. Again, I don't think you want to do that.

*   For searches and help try:

© Copyright 1996–2016 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index