 Konvergent discriminantialitys ovat yksi asioita, joita näet efter faktora-analysisen. Mitä on todella konvergentialitys ja discriminantialitys? Se on vähän ympäristöinen, koska nämä konseptit ovat olleet ympäristössä. Konverentialitys ja discriminantialitys muistavat kaikkiaan Kampbell ja Fiskit vuonna 1955. In that article, the authors presented a concept of multi-trade, multi-method matrix, which is a set of correlations when you measure at least two traits each with at least two methods. And they argued that some correlations in the table and lack of correlations are evident of convergent validity and discriminant validity without really explaining what exactly convergent validity means beyond the empirical criteria and the same with discriminant validity. So the current meaning of these terms has somehow evolved from the original meaning, because we don't really do multi-trade, multi-method matrices. It is very difficult to get two independent measures of the same construct. For example, measure two construct with paper and pen first and then with online form next. So that's not a practical way of measuring for most people. So the current practice in discriminant and convergent validity is that discriminant validity refers to whether or two scales measure distinct constructs. If two scales measure the same construct, they don't have discriminant validity. Then convergent validity refers to whether indicators measure the same construct. So it's kind of like used as a synonym for internal consistency. These are used also very inconsistently. So the definitions are a bit difficult to understand people use. Then without the definition, they are defined in different ways in the literature and they're also applied differently. This is a good example from an article by Carlson. And they say that some authors argue that a correlation as low as 0.28 of two indicators indicate convergent validity. So if two indicators are correlated at 0.28, then they measure the same thing. They also point out that some authors argue that correlation as high as 0.75 between two indicators is still not high enough to claim that these indicators are not distinct. So the current practice with these terms is that you have a number and then you make a claim. And there's really no connection of how high the number is. And because the discriminant validity and convergent validity concepts are not well defined, then it is very difficult to interpret what that correlation means. So you have to have an interpretation for the statistic. For composite reliability for alpha, you have an interpretation that it's a reliability of a sum of indicators. For these correlations between indicators, it's very difficult to interpret them directly. Therefore my take on this is that it's important to assess whether your scales measure distinct things. If you do a factor analysis, each indicator should load on one factor and two scales that are supposed to measure two distinct things should be two factors. If that is the case, then you have evidence of both discriminant validity and convergent validity and you don't really have to look at any of these correlations.