entropyetc - interpreation of measurements

[Ural Tuna]

Hi Statlist,
I have a survey data that contains the variables for measure attitude (typically in Likert scale). My main interest is to measure the degree of agreement/disagreement of the participant on the propositions. I thought that entropy measures could be fine for this aim, as I’ve read. So, I come across the program entropyetc which perfectly fits this job. I am sorry that I am technical-illiterate. My questions are:

1. Which measure in entropyetc is the best fit for this objective?
2. How can I interpret the results for each measure?

Thank you for any advice or comment.
(I use Stata 14)

Best Regards,
Ural

[Nick Cox]

None of the measures in entropyetc has anything to do with measuring agreement.

Perhaps you are confusing entropyetc and kappaetc. Both are from SSC. daniel klein is the author of kappaetc.

[Ural Tuna]

Thank you for the prompt response and comment. I am going to examine kappaetc right now.

Best wishes,
Ural

[Mike Lacy]

I think Ural wants a measure of consensus/agreement (un-dispersion) for categorical variables, ordinal ones in particular. (This is a different sense of agreement than what is relevant to -kappaetc-). If I'm understanding correctly, then I would suggest the ordinal measure of dispersion implemented in my -ordvar- , which is available at ssc. I quote shamelessly from -ssc describe ordvar-

. ssc describe ordvar
 ------------------------------------------------------------------------------------------------------------------------------------------------------
package ordvar from http://fmwww.bc.edu/repec/bocode/o
------------------------------------------------------------------------------------------------------------------------------------------------------
TITLE
      'ORDVAR': module to calculate measures of ordinal consensus and dispersion

DESCRIPTION/AUTHOR(S)
      ordvar calculates measures of ordinal consensus  and dispersion.
      These include lsq and 1-lsq, which are 0/1 normed ordinal
      consensus and  dispersion statistics described in Blair and Lacy
      (2000)......

[Ural Tuna]

Dear Professor Mike Lacy

I am grateful for your comment. Indeed, I’ve realized that kappaetcis related to different interest from mine. I guess I could not express myself properly before. My data contains some propositions (mostly on basic assumptions of economics such as the homoeconomicus argument) asked economists. Responses are mostly on Likert-type scales (answers at 10-scale is required in a few questions). I would like to know how much consensus there is among economists on each proposition. Later, I would like to compare economists by some demographics on their consensus level.
I have examined your ordvarand read your illuminating papers (Blair and Lacy 2000; Lacy 2006) cited in the help document. As I understand from Blair & Lacy 2000, it fits my research problem. I would like to ask that if I understand correctly, the normed dispersion takes a value between 0 and 1. As it goes to 1, the dispersion increases. Is that true? My second question is that will it work for 10-scale questions without any problem?
Finally, I will be grateful for any further advice on the method for my research problem.
I am thankful for your comments and works that help me.
Best regards,
Ural

[Mike Lacy]

Yes, the normed dispersion value ranges between 0 and 1. Between the various articles, I changed my preferred terminology a bit, which would confuse things. So, to clarify, using the terminology of the output of the -ordvar- command: The normed dispersion value, "1-lsq," ranges from 0, minimal possible dispersion or maximum consensus, whichever way you want to describe it. A value of 1 means maximal dispersion, or minimum consensus, which occurs if 1/2 of the responses are in category 1 and half in the top category, with no responses in between. The "normed consensus" value in the output is just 1-normed dispersion.