Patrik,

In my opinion, using absolute values is not a good choice in your case. Number of unemployed people is heavily dependent on population size and this value tells you little about magnitude of unemployment. On the other hand, unemployment rate is robust to population size and is comparable between populations of various sizes.

With number used you will also possibly have problems with heteroscedasticity because error terms in your regression will probably tend to rise as the population size increases.

Hope it helped,

Blazej

Yes, if you suspect that outliers in your data will bias your results, standardizing your variables is inevitable. Standardizing your variables will do what a median regression would do. However, you will have to standardize all your variables, not independent variables only.

Similarly, if you suspect you are likely to face multi-collinearity, generate a correlation matrix and identify the variables which are highly correlated (r>0.5). To avoid multi-collinearity in your regression, you will avoid including the correlated variables in the same regression.

Thank you, Eliud Moyi, this will help me a lot [...]

However, I am a geographer and my statistical background is not that good, maybe standardize is not the right terms to refers what I want. I meant normalize. For example, instead of use the absolute value of unemployment, maybe I should divide that value to the total population of that specific zone. This might help to avoid the such zones with higher population will not necessarily have a higher unemployment. Because a have difference in the size of zones and consequently difference in their population.

Thanks again,

Thank you, Blazej Lyszczarz...

It will help for sure,

Patrik. I get your problem now. The term you should have used is not standardizing your variables because this means something very different. Perhaps, the right terminology you should have used is scaling your variables. When you use a variable like unemployment/population, then you have scaled your variable. I wish you the best.

Yes you are right, Eliud! However, in Cartography we call it normalization also.

Thank you so much,

Hope you the best as well.

Dear Dr Eliud Moyi, in your answer to Mr Patrik, you said, ''If you suspect that outliers in your data willbias your results, standardizing your variales is inevitable. In this case, what are the likely outliers among his variables or data going by your judgement?

Sebastian, outliers in data are extreme observations in the data. They can only be determined by the person using the data. So, since I have not seen the database, I cannot tell whether there are outliers but Patrick can check the data.

Patrick. In economics, we use the term to normalize when our interest is to make the variable approach a normal distribution (be bell-shaped). The most common use of this is when we log-normalize household income to make it less skewed and approximate a normal distribution. We normalize variables because regression theory is built on the normality assumption - that the variables we are using have been sampled from a normally distributed population. A violation of this assumption has an adverse consequence on the regression estimates.

I guess that you have more than one problem. I would say that your major issue is tat you deal with a DV that is constrained in the sense that your DV cannot assume negative values (e.g. the absolute value of unemployed people cannot be negative). I would say that you should consider possible alternative approaches.

a) you can look at the changes in the value (e.g. if you have panel data);

b) you should consider the use of alternative econometric techniques that allow you for considering the constrains. For instance, you may want to consider Tobit regression.

Hope this helps.

Best regards

Andrea

Thank you Eliud.

Yes I know, you are right! I have read something about it.

Best Regards

Thank you Andrea Moro,

I appreciate it!

Best regards,