U can check the normality using the Kolmogorov Smirnov Test..

I.e if p value is statistically insignificant, then d sample selected is said to be normally distributed.

Thanks

Hope Osayantin Aifuwa

Aifuwa thanks...

My question was whether do I need to check the normality if my data set has 875 sample?

You may need to check, but Wooldridge (2013) a non-normality does not have a significant effect on the results of the ordinary Least Square regression in a relatively large sample.

Thanks Bala

Bala kindly attach the paper of Wooldridge (2013)

Yes, check it that is the right, because the normality is a condition for regression

OK. See attached. Check page 120, second to the last para as there are many discussions about normality.

Sorry I have uploading error, but I will send the book as soon as possible.

You must check your sample for very unusual observations (outliers) before, but normality is not nomally a big deal for regression techniques. After fitting the regression model, you have to check for homoscedasticity (equal variance), normality and independence of residuals, along with the absence of influential observations. The presence of influential observations (leverage and outliers) is the biggest problem when fitting a regression model. Check your assumptions prior to validate your model. R-sqr-predictive is one way to assess influential observations.

thanks for your reply @ Marquez

I had been trying to upload the book but I can't. You can download the book using the link below.

Jeffrey_M._Wooldridge_Introductory_Econometrics_A_Modern_Approach__2012.pdf

You need to check all the assumptions of regression such as normality, linear in parameter, homoscedasticity, collinearity etc.

For detailed procedure refer to Wooldridge text.

You can apply the Glejser test of Hetroscedasticity

Here are the links of the text book regarding

Jeffrey_M._Wooldridge_Introductory_Econometrics_A_Modern_Approach__2012.pdf

In Z library

https://b-ok.cc/s/?q=Wooldridge+Introductory+Econometrics+A+Modern+Approach+2012+

Be aware that in this universe Normal distribution is impossible and most response variables fall into other patterns of randomness. Usually, there also is a natural dependency between mean and variance of the response distribution.

Using hypothesis tests to check assumptions is a highly problematic practice. The preferred approach is to use a Generalized Linear Model and choose a matching response distribution upfront. That usually is straight-forward, see chapter 7 of my book draft:

https://www.researchgate.net/project/Book-New-statistics-for-the-design-researcher

@Martin, I guess I'm confused. If normal distributions are impossible in this universe why do you suppose that other continuous distributions are not only possible but easy to identify. It seems to me that simply asserting these things are true is not much of a reason to believe them.

I hope your book is more convincing than this. Is it? David Booth

David Eugene Booth: As I explain in my book, Normal distribution implies that the response variable spans from minus to plus infinity. There are no magnitudes in this universe that have neither a lower nor an upper bound. If you can name one, I would like to hear it.

Response distributions are skewed (not symmetric) and have less variance (not constant), the more they approach one boundary. In practice, Normal distribution is a reasonable approximation as long as responses stay far from the boundaries. For example, if in an experiment you instruct participants to work as fast as possible (and if they comply) you will probably get a more left skewed distribution of reaction time, because participants approach their limits.

This all has little to do with continuous vs discrete, which I regard a minor problem. However, it follows from quantum mechanics that nothing that is based on interaction of energy with mass can be truly continuous either.

30 is enough to consider that data is normally distributed according to the central tendency theory

Wafaa Salah: Your reference to the CLT is misleading in that it confuses the distribution of response variables (which was the question) with the distribution of sampling means. The CLT is only about the latter.

I promise you that if you take 30 (or 875) observations from a variable that follows Poisson (or Gamma, Binomial, ...) distribution, it will not become Normal all of a sudden. If you take 30 samples and calculate the sample average, this may well get close to Normal distribution.