Coefficients that should be negative

#1
Hi All

Please bear with me here. I am new and I also am only learning about statistics & regression analysis.

I have an issue where the data I have entered comes out as positive coefficients where they should be negative. The data in question is "underemployment" (in Australia). Generally speaking as underemployment rises it is not good for the economy as there is spare capacity in the labour market. However the data I have starts from say 2% and rises up to 8%, but the regression model picks this up as a positive factor.

How do I go about changing this to get the model to recognise this data as a negative factor ?

Thanks in advance
 

Karabiner

TS Contributor
#2
I suppose that it depends on whether there are additional variables in the model,
and how they are related to underemployment and to the dependent
variable, and how exactely your dependent variable is defined, and
whether "underemployment is not good for the economy" is a sound
theory at all.

With kind regards

Karabiner
 
#3
The Y variable is Automotive Sales

There are other X variables (GDP, unemployment, wages, CPI, etc).

I get the same with wages as well ... Logic is wages go up theoretically higher disposable income ... however it treats this variable as a negative factor.
 

noetsi

Fortran must die
#4
So one thing to do is go back to your source data and see if you coded it wrong somehow to generate this error. In logistic regression, which I don't think you are using, some softwares maximize 0 rather than 1 so you have to change the default (SAS does this). Also look at your p value and slope (effect size). If the p value is way above .05 or the effect size is very small it could be the sign is nonsensical (does not really impart meaningful results because you found a very minor or incorrect effect). Check to see, say with partial slope residuals, if your underemployment relationship is non-linear (your slope will be incorrect if it is or not mean what you think it does).

If that is not true than your model is saying the theory is wrong. Generally this is a good thing for researchers.... they prefer to find what was believed in the past was wrong...it helps to get published:)

If you want to force the model to generate the results you want, generally a very very bad idea, you could probably multiply the variables in question by -1 and get this. But again, you should not do this.