Brian's comment concerning the advantages of Krippendorff's alpha might not
be quite accurate. Although the original author implies that his coefficient is somehow superior to others in the literature, Gwet (2011, 2014) shows that (i) Krippendorff's alpha can be computed in a very similar fashion to other coefficients and (ii) other coefficients can be modified to handle ordinal, interval, ratio (, circular, bipoloar, ...) ratings, just like Krippendorff's alpha can. The method for handling missing data can also be extended to other coefficients. Details are given in Gwet (2014) and are implemented in -kappaetc- for Stata (code is completely open source). I would be interested in details about the "Krippendorf macros" that Art Kendall mentions, especially with respect to complex designs. Best Daniel Gwet, K. L. (2014). Handbook of Inter-Rater Reliability. Gaithersburg, MD: Advanced Analytics, LLC. Gwet, K. L. (2011). On The Krippendorff’s Alpha Coefficient. Manuscript submitted for publication in Communication Methods and Measures (August 2011). https://www.researchgate.net/publication/267823285_On_Krippendorff's_Alpha_Coefficient -- Sent from: http://spssx-discussion.1045642.n5.nabble.com/ ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
Daniel,
Thanks for chiming in. Actually I've never been a Krippendorff devotee. He really hasn't added much to Fleiss, and as he says, with greater numbers of raters and items, alpha and kappa are the same. I almost never use ordinal, interval, and certainly not ratio data for agreement. I just rely on Fleiss and Cohen (1973) on the equivalence of weighted kappa and the ICC. I've become more intrigued with approaches like Aikin or Ubersax (who advocates a latent class analysis approach.) I've also developed the approach that the choice of agreement stats depends on the nature of the data, and what we know. For example, are all categories equally probable, is there an underlying categorical structure that's not uniform, should raters be taken into consideration? I think each agreement study is different, and one size doesn't necessarily fit all. Brian ________________________________________ From: SPSSX(r) Discussion [[hidden email]] on behalf of daniel klein [[hidden email]] Sent: Friday, October 20, 2017 5:25 AM To: [hidden email] Subject: Re: SPSS Python Extension for Fleiss Kappa Brian's comment concerning the advantages of Krippendorff's alpha might not be quite accurate. Although the original author implies that his coefficient is somehow superior to others in the literature, Gwet (2011, 2014) shows that (i) Krippendorff's alpha can be computed in a very similar fashion to other coefficients and (ii) other coefficients can be modified to handle ordinal, interval, ratio (, circular, bipoloar, ...) ratings, just like Krippendorff's alpha can. The method for handling missing data can also be extended to other coefficients. Details are given in Gwet (2014) and are implemented in -kappaetc- for Stata (code is completely open source). I would be interested in details about the "Krippendorf macros" that Art Kendall mentions, especially with respect to complex designs. Best Daniel Gwet, K. L. (2014). Handbook of Inter-Rater Reliability. Gaithersburg, MD: Advanced Analytics, LLC. Gwet, K. L. (2011). On The Krippendorff’s Alpha Coefficient. Manuscript submitted for publication in Communication Methods and Measures (August 2011). https://www.researchgate.net/publication/267823285_On_Krippendorff's_Alpha_Coefficient -- Sent from: http://spssx-discussion.1045642.n5.nabble.com/ ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
Free forum by Nabble | Edit this page |