Applying multiple regression results to new data SPSS

1
votes

My study involves testing infants at multiple time points using a number of behavioural tests.

I used a multiple regression to establish which behavioural variables could predict severity of outcome score at the end of the testing.

So, I tested infants at 14 months and used a regression model to identify whether scores from any of the behavioural variables could predict better or worse scores at 24 months.

my model works fine and 4 IV's significantly predicted severity of outcome. But now I want to use this regression model to see whether it can predict 24m outcomes from a new dataset with new infants - BUT they have been tested on the same IV's and DV's.

Does anyone know how I might go about doing this in SPSS?

Thank you!!

2
regressionspssprediction
I don't quite follow this. Can you provide more information about the variables & the model? Note that 'how do I do ___ in SPSS?' questions are off topic here, but there may be a statistical question we can help you with. - gung - Reinstate Monica
I ran a multiple regression using behavioural data from infants. The IV's were behavioural measures (all continuous) and I used scores from these measures to predict whether an infant had a higher or lower general ability score later in development. The model showed that the scores from 4 behavioural measures predicted ability score later on. I now have a second phase of infants that have been tested on the same measures and have later ability scores. I want to use the new data in the original regression model to see if the model can accurately predict ability scores of the new infants - frankralph1
and wanted to know if this was possible as i only use SPSS - frankralph1

2 Answers

3
votes

Actually, the easiest way is to save the regression model as an xml (pmml) file when you run it and then use the scoring wizard on Utilities to generate the scoring code. The scoring is actually done with a COMPUTE using the APPLYMODEL function, but the scoring wizard will generate this code for you.

3
votes

You need to take the parameters from one regression, and apply them to a new dataset. There is more than one way to do this in SPSS. Here's the easiest way (that I know of), but see @JKP's answer for another way.

You merge the two datasets, with a flag variable saying which dataset they were from. All the variables have the same name EXCEPT the outcome variable.

For the 24 month dataset, set the value of the outcome to missing.

Run your regression, and save the predicted values. Because the new dataset cases are missing on the outcome, they have no effect on the regression model. But if they are not missing on all of the predictors, they will get a predicted value. You can compare this predicted value with the actual values, for dataset 2.

This is somewhat easier in SAS, R, or Stata - as all of these easily store regression results and allow them to be applied to a new dataset. You can do that in SPSS using the ODS system, but it's fiddly.

In Stata, run your regression, load your new data, type:

predict predvalues

In SAS, use Proc Score (takes too long to explain, Google will help you).

In R, use:

 dataSet2$predvalues <- predict(RegModel, newdata=dataSet2)