I now develop multiple linear regression model and I want to check the model accuracy. One of my friend suggest me to use Chi-square test however i am not sure about it. First, based on one text book, it is written that the data used in Chi-square are either counts or frequencies measured across categories that may be measured on any scale. Through, this information, i thought Chi-square are not applicable for multiple linear regression due to data type (observed and predicted variable).
Furthermore, i found that it is better to use some accuracy measurement namely, Mean Absolute Error (MAE), the Mean Absolute Percentage Error (MAPE) and the Root Mean Square Error (RMSE).
Please give me some suggestion regarding to Chi-square test and other statistical test or measurement
First of all, what is your definition of "model accuracy" (model parameters estimation? model predictions for new observations? In terms of bias, of variance, of both?) and what is an acceptable value of this accuracy in your application.
Then you can think about how to test it...
I agree with Emmanuel, it is important to understand what you mean by "model accuracy". I suppose you could use a Chi-square test to test how close the predicted values from the estimated regression equation are to the actual values of the dependent variable y, by creating intervals for y and counting how many instances of the predicted values fall in the same intervals. However I don't see why you would do that, when there are other diagnostic statistics that allow you to test the significant/fit of the model.
Please check this book:
http://www.statisticalassociates.com/assumptions.pdf
I think it will give useful information.
- Badges
- Science topic