We wish to thank Kate Graham for her comments [1] on our paper [2] and to respond to the issues that she raised. To recapitulate, by applying the first-difference (FD) method we submitted the alcohol–violence association to a more stringent test for causality than those used in previous survey research. Thus, using two-wave panel data from a general population study of young people, we estimated how changes in the frequency of violence correlated with changes in drinking. Further, we examined whether this correlation was contingent upon suppressed anger. The results indicated that alcohol may indeed be related causally to violence, but that the effect of drinking was confined to individuals who tend to withhold their angry feelings. Graham describes our study as innovative, but she also questions our statistical approach as well as our conclusion regarding the moderating role of suppressed anger.

The FD-method and other methodological issues

The major advantage of the FD-method is that it eliminates the effects of time-invariant confounders, but as Graham rightly points out, this method also has some important limitations. However, most of these limitations work in the direction of increasing the risk of type 2 error (rejecting a true hypothesis), which must be weighed against the reduced risk of type 1 error (not rejecting a false hypothesis). In his instructive methodology book, Firebaugh [3] notes that the first-difference method is conspicuously under-utilized in social science and pleads for a wider use of it.

In response to some more technical queries, we wish to clarify that the log-transformation of our measures on alcohol use and violent behaviour was performed to linearize the associations between the two. There are certainly other transformations that could do this job, but logging has the advantage that it yields a highly intelligible effect measure (elasticity). Further, with regard to the items used to assess suppressed anger, Graham asks why we used factor scores rather than a sum score. Our answer is simple: factor scores are advantageous because each item contributes to the scale in proportion to its relation to the underlying dimension. Moreover, the fine-grained scale of the factor scores made it possible to divide our sample into three equal-sized groups, whereas the coarse summed scale yielded groups that differed substantially in numbers. It may also be noted that we used a sum score in the first version of our paper, yielding an identical pattern of findings to that observed when factor scores were applied instead.

Graham also asks why Pearson's r was used to estimate cross-sectional correlations while the associations between the change scores were reported as elasticities. However, we did in fact also report cross-sectional elasticities (in addition to the Pearson's r), noting that these were twice the size of the elasticities based on change scores.

The moderating role of suppressed anger

According to Graham, the suppressed anger explanation that we propose may not apply to alcohol-related violence among young people. Referring to qualitative studies of bar-room aggression, she suggests that violent behaviour in relation to drinking may occur because it is considered normative among young males—not because withheld feelings of anger are more likely to be behaviourally expressed under the influence of alcohol. Indeed, for some subgroups of young males, this may well be the case. However, our study was based on a general population sample and we did not focus specifically on male-to-male aggression in the context of on-premise drinking. Moreover, only a tiny fraction of all drinking events (including those of young males) involves violence, and we examined whether alcohol in combination with the disposition to suppress anger could account for individual differences in aggressive behaviour. For both males and females, our hypothesis of a moderating role of this anger dimension was supported.

Graham also suggests that the above-mentioned effect of suppressed anger may be due to other dimensions of anger or to an angry disposition more generally. However, we focused specifically upon suppressed anger because it was the theoretically more interesting one (e.g. fitting well into the disinhibition perspective), and allowed ourselves to do so because our analyses indicated that it was empirically distinctive from other indicators of anger. We still agree with Graham that a more convincing strategy would be to conduct comparable analyses for both suppressed and expressed anger. However, we do not agree that our findings are inconsistent with experimental research showing no moderating role of suppressed anger on the link between alcohol consumption and anger expression. Thus, the one and only study that Graham refers to in this context [4] used facial expression of anger as outcome measure, implying that its scope differed from that of our study.

Declarations of interest