Sumedha,

First, I'm guessing these are observed predictors. In my fields (psychology and public health), standardizing is not recommended, but centering (a mean of 0, but not transforming the standard deviation) is in many cases.

Standardizing is not recommended because, among other limitations, it introduces more sources of error, as the standard deviation by which you are dividing is only an estimate, not the population standard deviation.

Centering, though, is often helpful, but it will not, as you saw, influence the parameter estimates for the interaction effect. It does, though, alter the estimates and interpretation of the main effects. Consider that the main effect is the estimated effect when all other predictors are held to zero. If you're interested in the main effect of A, holding B to zero also holds the interaction term (A*zero) to zero, isolating the main effect of A. It is generally helpful if 0 is a meaningful value, such as the sample mean.

Hope this helps.

Pat

Sumedha,

First, I'm guessing these are observed predictors. In my fields (psychology and public health), standardizing is not recommended, but centering (a mean of 0, but not transforming the standard deviation) is in many cases.

Standardizing is not recommended because, among other limitations, it introduces more sources of error, as the standard deviation by which you are dividing is only an estimate, not the population standard deviation.

Centering, though, is often helpful, but it will not, as you saw, influence the parameter estimates for the interaction effect. It does, though, alter the estimates and interpretation of the main effects. Consider that the main effect is the estimated effect when all other predictors are held to zero. If you're interested in the main effect of A, holding B to zero also holds the interaction term (A*zero) to zero, isolating the main effect of A. It is generally helpful if 0 is a meaningful value, such as the sample mean.

Hope this helps.

Pat

You may  not have mixed scale type data set  so the results are same...or did you check the significance for both cases?..we standardize data for mixed scale type or highly skewed data..normality issues...In order to plot and compare two different scales  to test the the moderation effect you need to standardize data. I advise to take results for  standardized data only. Please use statistical tool package from Statwiki . You can down load, it  has video also..I hope this is helpful

http://statwiki.kolobkreations.com/wiki/Main_Page

The Standardized Values are the regression coefficients after being standardized into equal unit namely standard deviation. We could use these standardized values for comparison purpose. This is especially so if the original data values are measured using different scales.

I'm going to qualify Zainudin's answer. Standardizing only allows comparison between different variables if both are exactly as representative of the population, which is rarely the case in convenience samples, especially when the predictors are substantively different in content  (I don't think Sumedha has said the nature of her sample). 

In other words, since both standard deviations are affected by the sampling approach, they have to be affected in the exact same way in order for standardized regression coefficients to be comparable. I don't think there's any easy way around this problem.

Pat