topic Re: Fixed Effects and scatter plot in New SAS User

Fixed Effects and scatter plot

GKati — Fri, 01 Feb 2019 14:57:20 GMT

Hello,

I would like to create a basic scatter plot after a fixed effects regression. I am using PROC GLM for the FE regression. Best would be to add an effectplot line to the my code, as such:

proc glm data=individuals;
class year (ref='2010' ) age_gr (ref='0' );
absorb id;
model cost=year|type/ solution;
effectplot fit / obs;
run;

but effectplot doesn't seem to work with GLM. What are my other options?

Thank you.

Re: Fixed Effects and scatter plot

PaigeMiller — Fri, 01 Feb 2019 16:19:24 GMT

@GKati wrote:

Hello,

I would like to create a basic scatter plot after a fixed effects regression.

Scatter plot of what?

Have you looked at ODS Graphics to see if the plot you want is already in there?

Re: Fixed Effects and scatter plot

GKati — Fri, 01 Feb 2019 20:38:01 GMT

Hi @PaigeMiller :

I would like a scatterplot of cost vs type with a regression line and 95% CIs. Sort of like the graph attached.

I know that this can be generated by adding the line

effectplot fit / obs;

to my code above.

The problem is that I need to perform a Fixed Effects and using PROC GLM is the only way I know how to do that and PROC GLM doesn't support the effectplot command.

Re: Fixed Effects and scatter plot

PaigeMiller — Fri, 01 Feb 2019 23:19:07 GMT

So the example plot you show doesn't really fit the analysis you are doing, as far as I can see.

So in the scatterplot you want, please be specific. What is the X-axis variable? What is the Y-axis variable? What are the lines on the plot, relative to the variables you have?

Re: Fixed Effects and scatter plot

GKati — Sat, 02 Feb 2019 19:45:47 GMT

The X- axis variable should be type, the y-axis variable should be cost. What would be a more appropriate plot in your opinion?

Re: Fixed Effects and scatter plot

PaigeMiller — Sat, 02 Feb 2019 22:07:34 GMT

I don't know what a good plot is, you still haven't really described what you hope to see/learn here.

The problem with the EFFECTPLOT is that is only defined for a single X-variable, but your model has two X-variables, plus an ABSORB variable. You can't plot just one variable of the model, you have to determine what you want to do about the other variables in the model (plus the ABSORB variable) in order to get anything meaningful. When there's more than one X-variable, as you have, usually people don't plot the actual data on a scatterplot, they plot the model residuals as the y-variable on a scatterplot.