Cultural Accommodation and Language Priming

We are fully aware that there might be large differences between the various countries in the Anglophone cluster. However, on a world scale, they share many similarities. For the cultural dimensions that we studied, the Anglophone countries generally score very similar to each other and noticeably different from the Netherlands. Furthermore, in studies that cluster countries on cultural dimensions (e.g. Hofstede 1980; Ronen and Shenkar 1985; Ashkanasy et al. 2002), Anglophone countries are generally found in the same cluster.

In the game setup profits and losses were treated as equal, although we are aware of the fact that profits and losses may differ regarding their impact: “losses loom larger than corresponding gains” (Tversky and Kahneman 2000, p. 144).

Our data of games 3 to 5 reflect, in a way, a dyadic panel. After all, we have a cross-section of individuals that play interactively in fixed pairs over time. So, three types of correlations or interdependencies simultaneously affect the per-individual—per-round observation of the high or low price choice: (a) each individual i in each pair affects the decision made by the partner j (the dyadic nature); (b) the decision of each individual i in round t might be correlated with this individual i’s decision in round t-1 (the time series nature); and (c) each individual i’s decisions may reveal autocorrelation due to individual i’s time-invariant features (the cross-section nature). For such data, a ‘one size fits all’ estimation strategy is not available. In the analyses that will be reported below, we decided to control for (a) to (c) by adding Partner’s choice in previous round (a), Number of round (b), and a series of player variables ranging from Gender to Basic cooperative attitude (c). Alternatively, we could have opted for another estimation strategy. With correlated data as ours, the estimation strategy should, in the end, be determined by substantive reasons, based upon the question at hand (Zorn 2001). In our case, in principle, two candidates for alternative estimation strategies are the generalized estimating equation (GEE) and the random-effect model (REM) (a third option, the fixed-effects model (FEM) does not work in our case given our already large number of dummies and the rather extensive list of time-invariant covariates). As GEE requires a selection of one of three working correlation matrices and REM can be estimated with GLS or ML, this gives five alternative estimation strategies, in total. We refer to Zorn (2001) for an explanation of the pros and cons of both estimation strategies and their sub-variants. Here, it suffices to remark that the overall sign pattern of results is not affected by the choice of estimation strategy. However, significance levels are, as the significance of our main and interaction independent variables often drops below the standard threshold level of ten percent. The reason for this is the large degree of autocorrelation of choices at the level of an individual player. Because our theory does predict such a (treatment-dependent) autocorrelation to begin with, we believe that the results reported below imply a better fit to our central research question.

According to the World Values Survey (WVS), the proportion of religious affiliation in the Netherlands was 53% Catholic, Protestant and 43% non-religious (Guiso et al. 2003). In our survey, the proportion of non-religious respondents was 52%, whereas the proportion of Catholic, Protestant respondents was 45%. The slight discrepancy is not surprising, as our respondents are younger than the general survey population in the WVS.

We acknowledge that listing a choice of cooperative versus competitive behavior when playing with a visible counterpart could be different from actual behavior in the real world. However, we do argue that this experimental setting brings us closer to measuring behavior than previous studies have.

Here, more fundamental research is needed, particularly by adding brain measures to our experimental design. It might be that different languages activate different parts of the brain, or might activate similar parts of the brain differently, biasing behavior in one way or the other. As it stands, we cannot be sure whether our main language effect is the result of cultural priming alone or more deeper-level processes within the human brain as well. Note that our particularly strong interaction effect indicates that cultural accommodation does play a key role.

We would like to thank Professor Michael Bond for this suggestion.

The legally relevant one may be the one in Dutch. However, our argument concerns the company culture.