Critique of:

Hollis, K.L., M.J. Dumas, P. Singh, and P. Fakelman. 1995. Pavlovian Conditioning of Aggressive Behavior in Blue Gourami Fish: Winners Become Winners and Losers Stay Losers. Journal of Comparative Psychology. 109(2): 123-133.

The journal article describes the ability of fish to undergo conditioning in order to increase aggressive behavior. This study attempted to answer two questions: 1) Does Pavlovian edge provided by Pavlovian signaling actually result in a territorial victory? and 2) Does that same competitive edge result in an indirect, long term advantage for signaled males? In order to respond to such questions the experimenters set up an intricate experimental model and presented their results in the form of graphs and figures. From these results, the experimenters determined that Pavlovian signaling did indeed provide both a short term and long term aggressive advantage for blue Gourami fish.

Several problems can be identified with this study. The first of these problems deals with the biological significance. When creating a specific study, an experimenter must ask himself, ÊWhy is this important?Ë and once he has identified the appropriate answer, he must relay it to his audience. Nowhere in this article did the author state why Pavlovian training on Blue Gourami fish is significant. Therefore, the reader is forced to either assume that the study was irrelevant or speculate about why the study is important. The ability of speculation is further limited by the lack of background information that is offered about Blue Gourami behavior.

The second major problem involves the experimental model. For this experiment, an initial sample size of 36 fish was obtained from different local distributors. During the study, several fish were lost to illness and by the end, only 22 fish remained. With such a small heterogeneous sample it is very difficult to control for contaminating variables. Had the experimenters obtained a larger number of fish, all from the same distributor, some of the variation (i.e. genetic background, environmental background, etc.) could have been eliminated and results would have been more convincing. Also, the experiments were conducted for only twenty eight days. When trying to determine long-term effects of Pavlovian signaling, a more sensible approach would have been to prolong the experiment and monitor the fish behavior at certain time points to discover the optimal time range within which Pavlovian conditioning remains effective.

Another problem with experimental design lies within the number of trials and replicates that were used for this study. There is no mention anywhere in the article of a replicate experiment using a completely different population of fish. Also, only two trials of this experiment were conducted: one on day 29 and the other on day 32. These trials cannot technically be considered trials of the same experiment because they are testing different things. The purpose of having one or more replicates is to eliminate the possibility that this behavior is caused by a factor that is unique to one specific group of fish. The purpose of conducting one or more trials is to insure that the results obtained were not a chance occurrence, but rather they are direct result of the experimental conditions. Both of these precautions were overlooked in this study and therefore provide a weaker argument for the hypothesis.

The procedure with which the fish were tested was also confusing. First, the experimenters divided the sample into two halves: experimental and control. These groups were further subdivided into four groups: PAV, PAV-L, PAV-NL, and UNP. The acronyms translate into Palvovian, Pavlovian-light, Pavlovian-No Light and Unpaired respectively. All groups were subject to Pavlovian training, however they differed in the way they were tested. and-light groups benefited from a signal given before their encounter with the rival fish. -no light and Unpaired groups did not. Even after such careful subdividing and pairing, the experimenters overlooked an important testing group. This group is a control group where no conditioning or signaling was in anyway imposed on the fish. According to Figures 1, 2 and 3, the experimenters never paired unconditioned fish to each other, nor did they pair unconditioned fish to the various groups of conditioned fish. Had the experimenters established these additional pairings involving unconditioned fish, the increase in aggressiveness (with respect to natural levels) could have been measured much more accurately. Provided that the data obtained supported the hypothesis, such pairings would have inherently strengthened the evidence in favor of the effectiveness of conditioning on Blue Gourami fish.

The first set of results suggested that the presentation of an unconditioned stimulus had a similar effect on all groups of fish regardless of their training. ANOVA tests showed that there was no significant difference with p values ranging from 0.07 to 0.49. All four groups of conditioned fish responded differently to the conditioned stimulus. The differences were shown by an ANOVA test to be significant with p values hovering around the 0.01 mark. When looking at Figure 3 in the text, it is difficult to assess these p-values. There seems to be a definite difference between the aggressive reactions of the UNP group and all of the others, but there are several points of overlap between PAV, PAV-L, and PAV-NL. The graph does not seem consistent with the p-values and is somewhat misleading. Another problem with the figures is that they never show the differences of the aggressiveness of the four groups with respect to a control. For example, in Figure 4, PAV-L fish defeated their PAV-NL opponents 100% of the time. With this representation, conditioned stimuli seem to be very effective in increasing aggressiveness. However, what if a control group was to rival the PAV-NL group and defeat them 90% of the time. This would indicate a 10% increase in aggressiveness rather than a 100% increase. This kind of data would invite several different interpretations including that PAV-NL conditioning is actually responsible for decreasing aggressiveness rather than that PAV-L conditioning is responsible for increasing aggressiveness.

The author concluded that the data obtained completely supported his hypothesis, ergo conditioning was effective in providing short term and long term aggressive advantages. As I have discussed above, this conclusion is weakened by the numerous flaws in the experimental model and the misleading figures in the text. In order to make a stronger argument, I suggest that the author obtain a larger sample size, create more replicates and trials, and utilize the control group in a way that would enhance the validity of the argument.