How to test forecasting accuracy of regression model?

Click For Summary

Discussion Overview

The discussion centers around methods for testing the forecasting accuracy of a regression model, specifically in the context of a thesis project involving an 8-variable regression analysis conducted using SPSS. Participants explore various approaches to validate the model's predictive capabilities.

Discussion Character

  • Exploratory, Technical explanation, Homework-related

Main Points Raised

  • One participant inquires about testing the forecasting accuracy of their regression model and seeks guidance on how to present the results using SPSS.
  • Another participant suggests a method used in neural networks, where part of the data is reserved for testing accuracy, implying a similar approach could be applied to the regression model.
  • A third participant emphasizes the importance of reporting the "F" statistic for the overall regression and outlines a method for constructing a confidence interval for the forecast, including the calculation of the standard error.

Areas of Agreement / Disagreement

Participants present different methods and considerations for testing forecasting accuracy, indicating that multiple approaches are being discussed without a consensus on a single method.

Contextual Notes

Some limitations may include the need for additional data for testing, the dependency on the specific statistical methods employed, and the assumptions underlying the regression analysis.

Who May Find This Useful

This discussion may be useful for students and researchers working on regression analysis, particularly those interested in forecasting accuracy and validation techniques in statistical modeling.

dune2
Messages
5
Reaction score
0
Hey, I have just finished running all the regressions for my thesis and I now have a nice 8-variable regression with an r² if 0.53. Almost all my hypothezised variables are significant. I am now wodering if there is a possibility to somehow test the forecasting accuracy of this model? I am using SPSS, is there anything that comes to mind? I just want to say something like: here is the printout, all is significant. Here is the forecasting equation, coefficients 1 though 8 do blablabla, forcasting accuracy is X. Is that possible?
 
Physics news on Phys.org
Well you sound like you know more than I do but I know that when you are working with neural networks, you train the network on only part of the data, and you reserve the rest of the data to test the accuracy of the network--to see if the network correctly "predicts the past." Maybe you can see if your model correctly "predicts the past" if you have data available for testing that has not been input into your model.
 
You should first state the value of the "F" statistic for the overall regression.

It is possible to construct a confidence interval for the forecast; you first need to compute the standard error of the forecast. Then (lower bound, upper bound) = (Yhat - SEYhattc, Yhat + SEYhattc) where Yhat is the point forecast, SE is the standard error of Yhat, and tc is the critical t (= approx. 2).

See http://www.rand.org/pubs/papers/P4365/
 
Last edited:
Ah, thanks!
 

Similar threads

  • · Replies 8 ·
Replies
8
Views
3K
  • · Replies 13 ·
Replies
13
Views
5K
  • · Replies 30 ·
2
Replies
30
Views
5K
Replies
3
Views
3K
  • · Replies 4 ·
Replies
4
Views
2K
  • · Replies 23 ·
Replies
23
Views
4K
  • · Replies 2 ·
Replies
2
Views
2K
  • · Replies 3 ·
Replies
3
Views
2K
  • · Replies 5 ·
Replies
5
Views
2K
  • · Replies 1 ·
Replies
1
Views
2K