Setting effect directionality

[Karen Jay]

How can I demarcate the direction of a measure in my dataset? For example, I have one test where favourable effects are higher numbers (number of correct responses), whereas for another favourable effects are lower (a quicker time is a better score).

[Clyde Schechter]

What do you mean by "demarcate?" Also, are we talking about discrete or continuous variables here?

[Karen Jay]

Demarcate as in I need to make clear to Stata the favourable direction of the test, and these will be continuous variables.
Thank you!

[Clyde Schechter]

Sorry, but I still don't understand. Stata neither knows nor cares about favorability or unfavorability of anything. Favorability is not even a statistical concept. It is an attribute of the data that exists only in the mind of the beholder. In what way do you want your notion about favorability to affect what Stata does with your data? How would you want the results of an analysis to differ depending on this?

[Andrew Musau]

How can I demarcate the direction of a measure in my dataset? For example, I have one test where favourable effects are higher numbers (number of correct responses), whereas for another favourable effects are lower (a quicker time is a better score).

I will add the following to Clyde's advice: In your example, if you run a regression with the variables "score" and "time" and present the results, your readers are presumably smart enough to look at the signs and confidence intervals of the coefficients on these variables and determine whether they have a positive or negative effect on the outcome. You may do more harm than good by manipulating the data so as to, e.g., reverse the coefficient on time as it may not be very intuitive to think of it as a decreasing measure. So what is important is having your readers understand what each variable represents. Having said that, it is common to standardize variables that represent, e.g., perception indices. You may have an index that measures political rights and has scores within the range 0-100 where 0 is least democratic and 100 is most democratic. On the other hand, a rule of law index may take values in the range 1-400 where a score of 1 represents very high respect for the rule of law and 400 represents little or no respect for the rule of law. In such a case, you can transform the indices such that scores lie on the unit interval [0, 1]; where 0 is "bad" and 1 is "good". This is through straightforward linear transformations. For the former, you divide the political rights index by 100 whereas for the latter, the transformation is 1-((index-1)/399). So for the latter, you have both converted the scale such that scores lie within 0 and 1 and have reversed the scale.