ChrisSnijders

12/04/16 10:54

The Social Orientation Value (SVO) is a measurement of someone’s “social preferences”. Literally, it is meant to measure the extent to which you care about what others get. Often, It is measured through the “RING measure” (this involves 24 preference elicitations, see Liebrand 1984 or the details at the SVO page), or through a series of decomposed games (cf. van Lange et al., 1997). In this latter method, you typically get 9 comparisons and are classified as egoist, altruist or competitor if you choose consistent with that label at least 6 out of 9 times. With Jeroen Weesie I came up with a potential SVO measurement too. I wrote about that in a previous blog post.

A more recent type of measurement is the "SVO-slider measure" (see Murphy, Ackermann, and Handgraaf, 2011 - Measuring social value orientation). The SVO-slider measure is a variant of Liebrand's RING measure. A participant gets to choose between 9 self-other distributions of points, and gets to do this 6 times. The below picture shows the first two of these six (picture taken from the Murphy et al. paper).

The picture is a bit hard to read - sorry about that - but let us try to focus on the second slider. The second slider varies from (85,15) to (100,50) and consists of points on the line in between these two (rounded to integers). In fact, in the logic of the SVO-slider, the (85,15) is a typical choice for a "competitor" and the (100,50) a typical choice for an "individualist". Graphically, the task for a participant for this second slider comes down to choosing a preferred point on the orange line in the below graph (again, picture from Murphy et al.):Note that there are two other prototypical self-other payoffs in the figure: (50,100) for altruists and (85,85) for prosocials. The mean allocation for self is computed (S_mean) just like the mean allocation to other (O_mean). The SVO-value then equals arctan((0_mean-50)/(S_mean-50)). The angle you get is used, if wanted, to categorize the participants.

- altruists: 57.15 < angle
- prosocials: 22.45 < angle < 57.15
- individualists: -12.04 < angle < 22.45
- competitors: angle < -12.04

The SVO-slider-measure has pretty decent test-retest characteristics, seems to correlate reasonably well with behavior in experimental games, and correlates seriously less well with actual prosocial behavior. In this sense, it is not much different from existing ways to measure social value orientations, as far as I know.

Let us however try to consider a bit more closely what the measurement actually does, using a related and perhaps bit more careful mathematical logic. If a person's social value orientation (r) represents the extent to which someone cares about an other person's payoff, it should equal:

U = [payoff to self] + r . [payoff to other]Now let us assume that for the second slider measure, a participant prefers the payoff (87,19). This gives us eight inequalities to work with:

The first one, for instance, says that: If we repeat this exercise for the other 7, we findThat's nice, because it shows that if a participant chooses (87,19), then his or her social value orientation lies between -0.5 and -0.44. So far so good. However, the inequalities that go with a certain choice could also end up to be inconsistent. If a participant prefers (89,24), we get: and these cannot all be true simultaneously.

This same exercise can of course be carried out for all of 9 possible preferences that a participant might have. In the enclosed file, I show the results of this exercise. That is, for each possible choice within 1 of the 6 sliders, I supply the implied upper and lower bound for the social value orientation (or show that it is inconsistent). Below you see part of this file: If you look at the second slider measure in the row with (89,24) you indeed see that the result is inconsistent. For the same slider, if you prefer (87,19) you find the [-0.50,-0.44] interval. There are cases, for instance in the first slider, where the lower and upper bound of the interval coincide. Consider the result for the first slider. Because in the first slider the payoff to self equals 85 for all choices, the inequalities that you get for a given preference are either r<0 or r>0. In one case, when the participant prefers (85,15), all inequalities are "r<0". In one case, when the participant prefers (85,85), all inequalities are "r>0". Both choices are consistent, even though what they bring us is not very restrictive. The other preferences imply both r<0 and r>0, which cannot be true.

The question is then how a standard interpretation of the SVO-slider-measure compares with the mathematical interpretation as suggested here (and also, whether the social orientation measurement that Jeroen and I came up with, leads to results that are different). To be continued ...

SVO-slider-results (CS v2)