That old SVO idea might not be so bad …

That old SVO idea might not be so bad …

I have written about measuring a person’s Social Value Orientation (SVO) a couple times before (see here and here). In particular, Jeroen Weesie and I came up with a single question SVO measurement that we thought might be more efficient, includes an internal test of transitivity, and correlates with existing (much more tedious) measures. I briefly reiterate the question we suggested:

We present you a situation in which you have to choose between 4 possible payments to you and a random other participant in this survey:

Carefully consider these options. Put a [1] behind the option you prefer. Then write a [4] behind the option that you least prefer. Now consider the two options that are left and put a [2] behind the one you prefer the most of these two. Now put a [3] behind the option that is left over.

1) 69 for you, 24 for the other [ ]

2) 61 for you, 55 for the other [ ]

3) 78 for you, 13 for the other [ ]

4) 64 for you, 34 for the other [ ]

If you consider the numbers that people write as a pattern (for instance: [3214]), then you can connect the patterns to an SVO-value that is consistent with a definition of utility for a given (self,other) combination as:

U (self, other) = self + svo*other

The following bit of Stata-code could be used to create the SVO-values (as suggested by Stefan I label the SVO-value “swo” for Snijders-Weesie-Orientation, modest as I am):

gen swo = .

replace swo = 0.07 if pattern==”2413″

replace swo = 0.20 if pattern==”2314″

replace swo = 0.33 if pattern==”3214″

replace swo = 0.45 if pattern==”3124″

replace swo = 0.58 if pattern==”4123″

replace swo = 0.73 if pattern==”4132″

replace swo = 1.00 if pattern==”3142″

The last line is a bit arbitrary, as strictly speaking you only know that swo > 0.82, but ok. NB All other patterns are not consistent with the abovementioned utility interpretation.

A while ago, I heard that Stefan Wehrli was running SVO-related studies on MTurk, and asked him whether he could include this single question SVO measure in a survey with also the SVO-slider measure. I wrote about that slider measure in a previous post because it has some weird properties that apparently nobody bothered about before, but that is not the issue today. The good news is that Stefan included SWO and SVO-slider in two waves of the survey. Great! The results are pretty decent too!

We see that the SWO-values are inconsistent in about 12-13% of the cases. These might be participants with truly inconsistent preferences or participants who have not been filling out the survey seriously (or both), and it is also in line with earlier studies that have use the RING-measure or the Decomposed Games method. Furthermore, the SWO-value that is most common is 0.07 (perhaps we should code these as 0 instead), with 43 and 44 percent in waves 1 and 2. Then the SWO-value of 1, with 31 and 33 percent. About 15 percent of the participants has values in between the extremes.

Correlations of SWO with the SVO-slider value are 0.72 and 0.73 (calculated over the cases where we have a SWO-value). Not bad at all, especially given that a test-retest of any SVO-value is likely about equally high. Now let’s see if I can get Stefan to actually write something about this together …