topic How to test whether the differences between two regression models are significant? in Statistical Procedures

How to test whether the differences between two regression models are significant?

hua — Wed, 30 Sep 2015 20:32:38 GMT

Does anyone know how to test whether the differences between two regression models are significant?

For example, I have the dataset as follow:

station year month day Tavg

1 2014 1 1 2.3

1 2014 1 2 3.3

1 2014 1 3 1.3

1 2014 1 4 -0.6

1 2014 1 5 -0.9

1 2014 1 6 2.1

1 2014 1 7 1.8

1 2014 1 8 1.3

1 2014 1 9 2.2

1 2014 1 10 3.2

...

1 2015 1 1 3.2

1 2015 1 2 5.3

1 2015 1 3 2.3

1 2015 1 4 1.4

1 2015 1 5 -2.0

1 2015 1 6 -1.5

1 2015 1 7 0.1

1 2015 1 8 2.3

1 2015 1 9 4.7

1 2015 1 10 5.8

...

2 2014 1 1 6.7

...

First, i did a proc reg for each station and each year(model: Tavg= year), so I have one regression model for each year and each station. Then I want to compare the difference between two regression models under one station to get if it is significant difference between slopes or intercepts. Has anyone know how can I perform this?

Thank you for helps!

Re: How to test whether the differences between two regression models are significant?

Haikuo — Wed, 30 Sep 2015 21:20:03 GMT

This seems to be relevant. http://www.ats.ucla.edu/stat/sas/faq/compreg2.htm

Re: How to test whether the differences between two regression models are significant?

lvm — Thu, 01 Oct 2015 13:33:46 GMT

proc mixed ;

class station;

model Tavg = station year year*station / s;

run;

Station effect tells you if the intercepts are different for each station. Year*station tells you if the slopes are different.

To get the unique intercepts and slopes, assuming significance, use

model Tavg = station year*station / noint s;