I have a dataset and I intend to use multiple linear regression analysis with excel since the dataset isn't that massive in volume, just about 500 rows. I have continuous numerical data and I already checked the scatterplots for linearity. Here are my questions:

  1. I applied the model I came up with to the dataset to compare the actual and predicted values. How do I know the range of the acceptable errors?
  2. How do I know if my model is already the best model?
  3. Do I have to remove outliers?
  4. How do I do the hypothesis testing for this?

Sorry, this is for my dissertation and I'm currently on break so I couldn't bother my statistics professor. Please just give me a walkthrough, I'm not sure where to start here.

  • $\begingroup$ you should try to ask only one question per thread $\endgroup$
    – Antoine
    Dec 12, 2016 at 13:22
  • $\begingroup$ @Wes, the fact that this is for the OP's dissertation doesn't really make this [self-study]. That basically means homework from a course or an exercise from a textbook, or similar. In addition, we don't usually add the [self-study] tag for the OP, but ask them to add it for themselves & read its wiki (there are sample comments here). That way, it is more likely the OP will be familiar w/ out policies, & adding the tag can indicate that they intend to abide by them. $\endgroup$ Dec 12, 2016 at 17:41
  • $\begingroup$ Apologies, still new to the system $\endgroup$
    – Wes
    Dec 13, 2016 at 9:18

1 Answer 1


One question per thread is best, but in short:

  1. To get an overall impression or idea, you could plot the regression line with confidence interval?

  2. Using forwards/backwards selection would potentially be a good way to construct the model and find out.

  3. Do you observe any individual values which are so extreme as to influence the overall results? There are plots and tests for this.

  4. Look into t.tests, Anova, chi-square and F-tests.

This is just an idea of what to consider and research, as answering all these questions in 1 reply would be quite an undertaking. Nothing mentioned above is exact/ exhaustive, but it should provide a platform for developing your approach to the analysis. Plus working it all out is part of the fun/idea of working on a piece of research.

Good luck with the dissertation :)

  • 1
    $\begingroup$ Re (2), forwards/backwards selection is not a good way to construct a model (see here: Algorithms for automatic model selection). $\endgroup$ Dec 12, 2016 at 17:43
  • $\begingroup$ Quite right, but the answer is intended as a place to start. I usually include variables based upon theory, and only remove if there are problems with collinearity etc. The most important thing is that some thought goes into how you arrive at your model. $\endgroup$
    – Wes
    Dec 13, 2016 at 9:16

Not the answer you're looking for? Browse other questions tagged or ask your own question.