How to recode and combine items on a 7-point Likert-scale dataset

[Chris Fotiadis]

For my thesis, i was handed a dataset with around 100 7-point Likert-scale items. My supervisor can not help me and my fellow students with Stata, since she is trained with SPSS, and neither Google search nor the manual granted any helpful results, so that is why i came to this forum.

I need help with two things.

1) I need to recode some of these items, as they are negatively phrased (for example some say i like working here, others say i would take an other job if i could). I need to recode them so the values display the same results.

2) I need to combine some of these items into one variable. There are six items referring to organizational commitment, and for correlations i need to combine these 6 items into one single variable.

Does anyone have experience with this subject, or has knowledge of a better place to use for help (in the future)?

Much appreciated.

[Nick Cox]

2) could mean many different things. Combining could be taking a mean or median, or something else. There is no reason why your supervisor should not be able to help there, as what to do is independent of how you do it in any statistical language. If you can elicit a precise definition then we can (probably) give you Stata code.

1) is a matter of reversing a scale. Some might want to point you to recode here but I would want to underline that

gen myvar2 = 8 - myvar

flips 1 to 7 and vice versa and correspondingly flips in between. Any value labels will need fixing.

But see Daniel Klein's revv from SSC, which does all this for you.

[Chris Fotiadis]

2) could mean many different things. Combining could be taking a mean or median, or something else. There is no reason why your supervisor should not be able to help there, as what to do is independent of how you do it in any statistical language. If you can elicit a precise definition then we can (probably) give you Stata code.

1) is a matter of reversing a scale. Some might want to point you to recode here but I would want to underline that

gen myvar2 = 8 - myvar

flips 1 to 7 and vice versa and correspondingly flips in between. Any value labels will need fixing.

But see Daniel Klein's revv from SSC, which does all this for you.

Thank you very much for your quick response. I have immedietly tried your suggestion from problem 1) and it seems to do the trick. Problem 2) I may not have formulated precisely enough. There are 6 items for organizational commitment and 8 items for performance. I want to know the correlation between organizational commitment and performance. That is what i meant with to combine it in one variable. I dont necessarily need the variable itself, but I do need to know the correlation between everything that counts as commitment and everything that counts as performance.

[Nick Cox]

The only way I understand that is that you want a correlation matrix for commitment variables and performance variables, which you can get from whatever correlation command you deem suitable. If you seek a single correlation between commitment and performance without an arbitrary reduction to summary or composite variables, I think you need to spell out how that would be calculated defensibly.

[daniel klein]

1) revv is a somewhat older version of rev, but both should work

2) might call for (confermatory) factor analysis (see sem)

Best
Daniel

[Phil Bromiley]

You should also look at alpha which also deals with construction of scales.