Depending on your ratings, the kappa2 command can give you NaN for Z value and p value. For more explanation, see Cohen`s Kappa. If you receive NaN, it is best to completely omit the Z and p values, perhaps with a note that they could not be appreciated for your data. As you can see, this showed much better results: 97% approval and a cappa cohen of 0.95. Depending on your reviews, you can get a value for Kappa that is zero or even negative. For more explanation, see Cohen`s Kappa. Calculated estimate of points from Cohen`s kappa statistics. cohen.kappa can use either the weighting of the resemblance (diagonal -0) or the weighting of diversity (diagonal – 1) to match different published examples. A 50% agreement is much more impressive if there are six options, for example.

In this case, imagine that the two spleens roll a cube. Once in 6, they would get the same number. So a percentage of agreement by chance, if there are six options, it is 1/6 — about 17% approval. If two advisors agree 50% of the time, if they use six options, that level of agreement is much higher than we would have foreseen by chance. So you want to make sure that the degree to which two programmers agree on coded results is high. A simple way to do this is to consider the Interrater agreement: the number of votes that councillors agree on, divided by the number of things that are evaluated. The problem is that if the spleens work with a codebook from which a limited number of categories are available, they probably only agree by chance. Even a stopped clock is correct twice a day, and even inexperienced advisors who code things will eventually agree. Indeed, many of the things we want to do in the field of research will only happen by chance. To be a good researcher is to ensure that the things that happen in our research are probably not due to chance. Cohens Kappa corrects this by taking into account the number of times advisors agree to make random decisions.

Cohen`s Kappa can be used for two categorical variables, which can be either two nominal variables or two ordinal variables. There are also other variations: one of the problems with the percentage agreement measure is that people sometimes agree by chance. Imagine z.B. your coding system has only two options (z.B “level 0” or “level 1”). Where there are two options, but by chance, we would expect your agreement as a percentage to be about 50%. Imagine, for example, that each participant pours a coin for each participant and encodes the answer as “level 0” when the coin lands heads, and “level 1” when it lands in the tail. 25% of the time both coins will come heads, and 25% of the time the two coins would come dicks. In 50% of cases, councillors would therefore only agree by chance. So a 50% deal is not very impressive if there are two options.

This interpretation, however, makes it possible to describe very little correspondence between the advisors as “substantial”.