I'm having trouble creating a composite measure. I have 5 different non-normal variables (all different scales). I'd like to create one score that takes into account these 5 variables. I can manually weight the variables. How do i do this? I have JMP, so I have been trying to use that, but am not sure where to begin. All the of the variables have different scales and units.
Solved – Creating a composite measure
compositejmpnormalizationz-score
Related Solutions
General points on composites using t-scores
By t-scores, I assume you are saying that you have four variables each of which have been standardised so that the mean is 50 and the standard deviation is 10.
By using standardised scores (i.e., t-scores) as your component items, you are in some sense ensuring that the weighting of each variable in your composite is equal. This is often a desirable property where you believe conceptually that each variable deserves equal weight in the composite.
One issue with using standardised scores when forming composites is that the means and standard deviations used to form the component t-scores can change. This can then prevent the comparability of total scores. Thus, I would recommend that if you do use t-scores as the component variables, make sure that the formulas that go from raw scores to t-scores don't change over time.
Another point is that the mean or sum of a t-score will not be a t-score, but you could restandarise the total score.
Justifications for indexes
More generally, there are several justifications for creating a composite including:
- Common construct: the four component variables correlate and are theorised to reflect a particular construct.
- Reflects a conceptual category: The composite reflects a conceptual entity made up of multiple elements that may not necessarily be correlated. For examples, measures of city liveability often include quite diverse indicators (e.g., air quality, crime statistics, etc.). Regardless of whether they correlate, they all tap into an important construct which is city liveability.
- Predictive index: In some cases an index is formed for its predictive value. For example, a set of tests might be combined to predict job performance. In such a case, the weighting of component variables would often be influenced by a predictive model (e.g., regression coefficients in multiple regression).
These justifications also map broadly onto discussion of reflective and formative indicators (Diamantopoulos & Winklhofer, 2001).
Reliability and validity
All of the above justifications for indexes relate in various ways to reliability and validity. In general, I think most composites in psychology are created because they are believed to reflect a common construct. Thus, we combine items together to measure depression or extraversion or well-being.
I think you need pretty strong justification for combining items that don't correlate and calling that composite something psychologically meaningful.
Both reliability and validity have many meanings and nuances.
There are internal measures of reliability which will use the intercorrelation of component variables to estimate internal consistency reliability (e.g., alpha). That said, if you combine several items that have good test-restest reliability but to each other are uncorrelated, you will still get an overall scale with good test-retest reliability.
Validity means various things. In a broad sense, validity pertains to whether the inferences you want to make from the test are valid. Thus, if you want to use the measure to predict something useful then it's predictive capacity will be relevant issue. If you are trying to accurately reflect a meaningful integrated construct with some external existence then you would probably be looking for something where component variables intercorrelate. In other cases, you may want to be aligning with existing theoretical conceptions of a construct.
Additional material
I also have a large number of notes on composite score formation here around z-scores.
References
- Diamantopoulos, A., & Winklhofer, H. M. (2001). Index construction with formative indicators: an alternative to scale development. Journal of Marketing research, 269-277. PDF
There are two ways you can take: (1) just use the sums of scores, (2) use an Item Response Theory (IRT) based method. Using sums of raw scores is very common in social sciences but many psychometricians do not consider it being a sound approach. If you sum up the different questions from the questionnaire you assume that every answer provides you with the same amount of information - and in the real life that is not true. However, your data provides you in information on both the "abilities" of your responders and on precision of your questions, so that you can use both sources of information to gain deeper understanding of both your questionnaire and your responders. This is a pretty wide topic so you can check different resources on this topic, e.g. here, here or in this book. IRT will let you to use your data to obtain information on latent features measured by the questionnaires on continuous $Normal(0, 1)$ scale, so it also makes life easier with further analysis. It is mostly used in the area of educational research, so don't get discouraged that most examples in the books and articles are on measuring student abilities, because the method could be used for analyzing any kind of test or questionnaire data to obtain the latent profiles of the responders.
There are many statistical packages for IRT, for example, in R you can use mirt or ltm.
Best Answer
In general, a simple weighted linear composite can be formed as follows:
where w1 to w5 are your five weights and x1 to x5 are your five variables.
The question is what weights should you use?
A common approach in my field (psychology) would be convert each variable to a z-score and then unit-weight the variables (i.e., take a simple sum of z-scores). This in some sense represents an equal weighting that controls for the fact that the variables are on different metrics.
Even if you want to weight some variables conceptually more (e.g., variable 1 is more important), it can still be useful to first convert the variables to z-scores and then apply differential weights based on your conceptual weighting.
In other cases, the variables are on roughly the same scale (i.e., very similar standard deviations). In which case, you can often skip the z-score step. You see this a lot with multi-item self-report scales (e.g., life satisfaction, personality, etc.).
Also, sometimes you have items that are negatively related to the construct and you need to reverse certain items. So in that cause, you could use a weight of -1 instead of 1 after z-score transformation.
I've received a few questions about this over the while (see here).
I've never used JMP. But most general purpose statistics sofwtare have tools to create new composites: A quick google suggested this might be useful:http://www.jmp.com/support/help/Formula_Editor.shtml