Using proportions rather than the absolute values of the independent variables cannot yield the same results. This is because of unequal denominators. I think the use of absolute values will convey the right interpretation.

Thank you dear Ette Etuk,

I agree with you,

But I think use absolute value will give me some high correlation between independent variable, because most of them come from population, and also the size of population vary to much across the study area. The study area has some zones that is very small and other that is very big, in size (both, area and population).

I think it's if the value is left in their natural unit. You can always express or explain things in percentage term after the solution is found.

Hello Patrik,

If you do a regression of Count on some other variable X, the regression coefficient will have units of counts/Xunits.

From this you can see that regression on a variable Nemployed will give you a different estimate (counts/count) than regression on %employed (counts/%). So I suggest you think about what you want to interpret: change in Counts with change Nemployed, or change in count relative to %employed.

It strikes me that the former will be easier to interpret.

As to correlation of Nhousehold with Nemployed recovered from Nhousehold, that is possible. But before trying to fix the problem see if it is a problem (look at the correlation). If the correlation is small, you can use a sequential analysis,

Count ~ Nhousehold + %employed

This gives dependence on Nhousehold (% employed)

followed by dependence of Count on %employed (adjusted for Nhousehold because it is listed first).

And as a caution, analysis of count data (bounded at zero) can result in violations of the assumption of normal errors, which is an important assumption for regression routines, unless you specify some other error structure besides normal.

~David S

Dear David S, Thank you very much. I really appreciate it.

However I just want to make sure I understood what you have said. When you refer to Count ~ Nhousehold + %employed , you are mention Nhouselhold and %Employed as two independent variables or you are mentioning what is called interactions terms (which is the combination between them?

Hope you understand me!

Thank you again!

Patrik

Thank You Emerson Abraham Jackson!

I will keep you feedback in mind.

Patrik

Dear Patrik

I wish to submit that the kind of problem in hand, i.e. crime and education, you may proceed with figures in percent as independent variables. Data on various economic aspects such as interest rate, GDP growth rate, inflation, fiscal deficit etc are most of the time taken in percentage in place of count/absolute values

Hello Patrik,

Regression routines require an dependent variable and one or more independent variables.

Count ~ Nhousehold + %employed

Count is the dependent variable.

Nhousehold and %employed are the independent variables.

There is no interaction term in the model above.

Regression routines produce sequential sum of squares estimates and adjusted sum of squares estimates. The description from me, above, was for sequential sum of squares. Which means that the estimate for first term (Nhousehold) is not adjusted for the second term (%employed); the estimate for the second term is adjusted for the first term. However, people tend to use adjusted sum of squares estimates. Which means that the estimates for each term are adjusted for all other terms.

So now you get to think about what makes sense for you.

Do you want to interpret relative to Nemployed or relative to %employed?

If you decide to use both Nhousehold and % employed in one model, you get to decide what makes sense for you--sequential or adjusted.

As you have seen there are a variety of answers to the first question. And similarly, there will be a variety of answers for the 2nd question, sequential or adjusted. The correct answer is what makes sense for you, for this problem, and perhaps also based on what is usually published.

~David S

Thank You Bhag S Bodla.

It helps.

Thank you David Schneider.

It helps!