Can you explain why you want to do this?

Technically it is easy to do - if they are continuous, I would calculate a z score so they all have mean of zero and a standard of deviation of 1. But in lots of situations I would keep the original data - so it would depend on purpose whether or not I would do this transformation.

Can you explain why you want to do this?

Technically it is easy to do - if they are continuous, I would calculate a z score so they all have mean of zero and a standard of deviation of 1. But in lots of situations I would keep the original data - so it would depend on purpose whether or not I would do this transformation.

I am compiling 10 micro institutional variables for the purpose of creating one unified macro variable (the compiling of the separate variables into one macro variable is a common method in what i'm working on).

And thank you for your input. I actually was able to standardize the variables through the z-score standardization option on SPSS. However, I do have one additional question Professor Jones, if I may. One of the micro variables was continuous and did not vary at all along the observations (meaning all the observations had the same score). As a result, a standardized z-score could not be computed. Any idea how I may address this issue?

Thanks!

As Kelvyn Jones suggested, the calculation of z-scores for every variable is an acceptable option.

In regards to your last question, the best option is to remove the variable that does not vary along the observations (low heterogeneity) from the model to build the composite index (or macro variable, as you called it).

Also, as part of the variable selection process to include in the model, I suggest to analyze the col-linearity among variables. If two variable are strongly correlated, it is enough to include just one of them in the model. This could be done just calculating a correlation matrix.

Thank you!

Eltion,

I agree with Kelvyn that you should carefully consider whether to standardize. It sounds as though it's a regular practice on your area. But you should be aware that it brings in limitations and sources of error.

That said, for your other question, that micro variable should be discarded, as Ramon says. To be flip, a variable that does not vary isn't a variable!

Pat

A couple more issues - I understand that you are trying to build a composite index - so

(1) make sure that the variables go in the same direction ,eg most positive values on ALL variablesmeans thay you have a lot of the underlying concept ;

2 if one of the variables is very skew it could have a dominant effect on the final variable .- you may waant to transform to a more symmetric distribution before summing.

This paper discusses some of the issue involved with an example

Martin D, Senior M L, Williams H C W L, 1994, "On measures of deprivation and the spatial allocation of resources for primary health care" Environment and Planning A 26(12) 1911 – 1929

Hello Eltion Meka, I'm not sure whether this answer is applicable to your scenario, but I did a meta-analysis of left ventricular hypertrophy and there were different methods of calculation. However at the end of the day to enable comparison, each criteria was able to be categorised as with or without the condition using a cut-ponit. Then you can categorise and compare as appropriate. Thanks Debbie

Thank you all for your input! Your responses have been very helpful.