When a research participants expectations produce the results of an experiment?

Implications

Three kinds of implications flow from the work on interpersonal self-fulfilling prophecies. The first are the methodological implications for the conduct of scientific inquiry to minimize the effects of experimenter expectancy effects (and other effects of the experimenter). For example, the use of double-blind research designs flows directly from the need to control a variety of experimenter effects, including the effects of experimenter expectancy. More indirect consequences of a methodological sort flowing from this research are some of the work focusing on newer procedures for the analyses of scientific data, including work on meta-analysis, on contrast analysis, and other developments in significance testing and effect size estimation.

A second line of implications involves those for the study of nonverbal communication. For some 35 years, there have been strong reasons to suspect that the mediation of interpersonal expectancy effects depends heavily on unintended nonverbal communication, and this has generated a good deal of work on that topic. A third line of implications involves those for the practical consequences of these phenomena—for example, in classrooms, clinics, corporations, courtrooms, and, in particular, interpersonal contexts, such as those involving gender and race.

All three types of implications have been investigated intensively, but much of what needs to be known is not yet known. It seems very likely, however, that efforts to fill these gaps in knowledge will be repaid by noticeable progress in the methodological and the substantive development of the sciences.

XI Experimenter effects

Experimenter effects are errors introduced during the collection or analysis of experimental data due to the behavior of the experimenter. They can affect the data collected in an experiment and thereby confound the analysis of results in at least four ways (Rosenthal, 1976) through: (1) subtle differences in participant treatment; (2) errors in recording data; (3) errors in selecting cases; and (4) errors in the analysis of data. In the vast majority of cases in which experimenter effects have been studied, these effects bias results in favor of the hypothesized results, with the experimenter unaware that he or she is even making an error or treating participants differently based on expectations.

Rosenthal (1976, 2005) suggests several strategies for addressing experimenter effects. The best known is the use of experimenters who are blind to the condition being run. We mentioned using blind interviewers previously with reference to assessing the extent to which scope conditions are not instantiated for a given subject. When an experimenter is blind to the condition, his or her expectations regarding the hypothesized outcome for the condition are controlled, and so too is any cuing behavior caused by such expectations. The careful development and assessment of standardized prerecorded instructions can have the same effect of controlling for differences in experimenter behavior since the behavior of the experimenter, shy of the manipulation of the independent variable, is the same for every participant.

Another technique for the control of experimenter effects is to minimize contact between the experimenter and the participant. The less exposure a subject has to experimenters, the less likely it is that cues from the experimenter will be “picked up” by the respondent. Prerecorded instructions greatly reduce the amount of contact between participants and experimenters. Because any errors or problems in a video recording will affect all participants, however, careful pretesting and development of these instructions to identify and rectify problems is that much more essential.

Researcher

Closely linked to demand characteristics of the setting are the expectancies and authority of the experimenter. Just as the laboratory researcher exquisitely controls the setting, so too does the researcher, if not monitored, control the participant. Stanley Milgram's famous social psychology study showed that laboratory participants would shock another participant (seemingly to dangerous levels), merely because the researcher in authority told them to do so. Experimenter effects, effects generated unwittingly by experimenters, have been a source of much thought and research for several decades, ever since Robert Rosenthal introduced the concept and a line of research on the topic.

A typical demonstration of the experimenter effect uses a number of participants as experimenters. Half of these are led subtly to believe that the participants will perform a certain way and the other half are given a different expectation. Almost invariably, the participants as a group respond in conformity with the expectation given by their “experimenters.” The actual experimenters were puzzled about how to explain this dependable and replicable finding. The central question concerned how an experimenter's expectation was transferred to the participants. Two groups of “experimenters” read to their participants exactly the same instructions, except, of course, for the word or words that made the instructions different for the two groups of participants. No one heard any difference in the readings. Finally, a colleague detected a slight difference in the emphasis given a key word. In a carefully designed and executed study by Duncan and Rosenthal, some student instructors were led to expect that the participants they would be working with would tend to make ratings toward the success end of a scale while other student instructors were led to believe that their participants would tend to make ratings toward the failure end of that scale. And that is the way the two groups of participants tipped their judgments. The researchers concluded that the way in which the experimenter reads instructions to participants, even when instructions are read accurately, can significantly determine the participants' responses to an experimental procedure.

Researchers can influence participants in many subtle ways. As demonstrated by Rosenthal's program of research, when experimenters know their hypotheses (which will be the case for all except hired staff), and they know the experimental condition or assessed selection score of the participants, they may inadvertently behave in ways that confirm their own hypotheses.

E Summary of Section X

Although “laboratory effects” in mouse behavioral phenotyping studies have been well documented, with the majority of the variability accounted for by the experimenter, there has been little attempt to analyze the behavior of the experimenters that causes this variability. This section has focused on the detection, analysis, and correction of experimenter errors. These include observer effects, errors of observation, observer bias, recording and computational errors. In addition, there can be differences in reaction time between experimenters, effects of experience, and differences in handling animals.

One of the main problems in the study of behavioral phenotyping, as discussed by Blizard et al. (2007) and Stanford (2007), is the lack of training in the behavioral part of behavioral neuroscience. Many behavioral experiments are conducted by researchers with no training in methods of measuring behavior. As pointed out by Blizard et al. (2007, p. 138), “the real problem is rather a lack of pure laboratory behavior scientists,” and “we need to go back and fix a key element—namely the intelligent control, measurement, and interpretation of behavior.”

The future of the gene–brain–behavior approach to mouse models of human disease may therefore be a gene–environment–brain–behavior model that will require much greater emphasis on the behavioral aspects of the experiment (Fagiolini et al., 2009). In this case, the application of ethological techniques for the study of behavior (Lehner, 1979; Martin and Bateson, 1993), combined with the experimental design and analytical procedures of classical comparative psychology and the neurobehavioral techniques of classical physiological psychology, will provide powerful tools for the neurobehavioral analysis of transgenic mouse models of human neural disorders.

2 Evidence That Nestling Calls can Alert Predators

It seems intuitively obvious that nestling sounds will attract acoustically orienting predators, but in fact that may not be true, given the varying sensory abilities of predators, the competing demands on their time and attention, and the availability of alternative prey. It is therefore necessary to test this possibility. Unfortunately, direct tests for a predation cost of begging are few, and those that have been done apply methods that limit their results in distinctive ways. Most fall short of quantifying the marginal costs of incremental increases in calling, even though these costs are a key feature of many theoretical models of begging behavior. Because the predation cost of begging is a key issue in understanding the risks of communication between offspring and parents, we discuss the methods used to tackle this issue, starting with correlational approaches using natural nests and then turning to experiments using playbacks of begging calls and increasingly realistic methods.

The correlational approach was first taken by Redondo and Castro (1992). They found that black-billed magpie (Pica pica) nests that had more begging activity, as measured by the number of nestlings gaping or calling, were more likely to be depredated. Also, predation occurred sooner the more nestlings begged, although this result was only significant for the number of gaping nestlings, while showing a trend for number of callers. Interestingly, the results also showed that among partially depredated nests, there was tendency for predators to take those nestlings that were the fastest to beg, although here no distinction between gaping and calling was made. Overall, this study provides some evidence for a predation cost to begging and also suggests that predation costs could be borne by individuals, without necessarily affecting the entire brood (Section II.A.1). It is possible, however, that the correlations between begging and predation may have been due to confounding variables. For example, perhaps parents at those depredated nests had poorer defensive abilities, or had not sufficiently concealed their nests (Moreno-Rueda, 2007). Also, begging intensity was measured at each nest by placing nestlings in a bag and stimulating begging with “a single human vocalization,” a method that might not accurately reflect begging to parents and might be vulnerable to experimenter effects.

Intervening variables might also be important in explaining the negative results of another correlational study on the relationship between nestling vocalizations and predation. Halupka (1998) found that meadow pipit (Anthus pratensis) broods that begged with calling when their nests were checked, and were thus presumed to have higher average begging levels overall, were no more likely to be depredated, and had similar survival times, to broods that begged less intensively. He further suggested that parental behavior, especially alarm calls, would normally alert nestlings to the presence of predators and reduce begging in their presence (Section III.A.1). He thus concluded that, in this case, begging calls did not carry a predation cost. One could equally conclude, however, that begging does have a predation cost, but the cost is mitigated by other factors, including parental behavior.

More direct evidence for a predation cost of begging calls comes from experiments using artificial nests with playbacks of nestling calls (Table II). Such experimental tests began with Haskell (1994), who played western bluebird (Sialia mexicana) begging calls next to artificial nests placed either on the ground or in trees at a single study site, and found elevated predation rates at nests playing begging calls on the ground, but not in tree nests. In a follow-up experiment, playing begging calls at different rates at ground nests, he showed that artificial nests with higher call rates were more likely to be depredated. These experiments have been criticized for being unnatural in several ways that may have increased the likelihood of finding an effect (Haskell, 2002; Moreno-Rueda, 2007). Most notably, western bluebirds are cavity nesters with highly detectable and localizable calls (Section II.C.1) that would increase the likelihood of eavesdropping, especially when the calls were played from open nests placed on the ground. Furthermore, the calls were played constantly throughout the day, rather than more intermittently as might be found in natural nests, and nests would not benefit from parental defense. These departures from realism were intentional however, and adopted to increase the likelihood of finding an effect if one existed (Haskell, 1994). Another potential limitation was that the results might have been attributable to a few individual predators, because each experiment was conducted at a single study site.

Table II. Playback Studies of the Predation Cost of Begging Calls

A playback experiment that removed some limitations of Haskell's (1994) study nonetheless found similar results. Leech and Leonard (1997) conducted a similar experiment on tree swallows but used several study sites to ensure that predators were not the same individuals. They also played back nestling calls from ground and elevated locations, except here the elevated locations were nest boxes identical to those used at nearby study sites. In a further attempt to mimic the natural situation, they broadcast the calls from within the nest boxes, but placed the bait eggs (quail eggs) on a platform outside the nest to avoid confounding signal detection with the predator's ability to take the eggs. Finally, they played calls only in the morning and evening, rather than throughout the day. With all these modifications they, too, found that nestling calls increased the rate of predation, but here in both ground and elevated nests (Fig. 1). Placing eggs outside the nest box might have led to overestimating absolute predation rates, but begging calls clearly increased the risk of predation.

Dearborn (1999) emulated natural conditions still further by including both artificial and natural nests in the same study to analyze the effects of interspecific brood parasitism on predation rates of parasitized indigo bunting (Passerina cyanea) nests. Songbird nests containing nestlings of the brood parasitic brown-headed cowbird (Molothrus ater) might be particularly subject to predation, because cowbird nestlings tend to give calls more loudly and at a higher rate than host nestlings, and because adults visit parasitized nests more frequently than unparasitized nests (Hannon et al., 2009). Dearborn (1999) found higher predation at nests playing back cowbird calls compared to silent controls, with playback of indigo bunting nestlings alone receiving intermediate levels of predation. Reassuringly, predation rates at the nests with playback were similar to those at the natural nests they simulated, making it less likely that the playback results were attributable to unnatural characteristics of the nests or playbacks.

Recently, McDonald et al. (2009) improved still further on previous experiments, by choosing more natural playback locations, more realistic delivery of calls, more realistic diurnal timing, and better replication of playback stimuli. At four colonies of bell miners (Manorina melanophrys), begging calls were played back from speakers placed by used nests, which in turn were placed in species-typical nest locations, in transects starting within the colony and continuing beyond it. Playbacks were made from hundreds of exemplars of begging calls, and were compared with playbacks of white noise modified to have the same amplitude envelope as the begging calls, thus providing a comparison to test whether the frequency structure of the begging calls reduced the likelihood of predation.

Begging playbacks increased predation rates compared to the silent controls, but their effect was no different from the modified white noise playbacks. While McDonald et al. (2009) suggest this implies that begging calls are not structured to reduce predation risk (as discussed further in Section II.C), it is also possible that both begging calls and white noise have acoustic features, such as energy dispersed across a wide frequency range, that minimize predation risk, and so predation rates did not differ for that reason. Curiously, predation rates did not differ between nests in and out of the colony, despite adult bell miners being vigorous defenders of nests, clearly a result that requires further study.

Overall, these playback experiments are important because they show that begging calls can attract predators to nests, but they provide relatively little information on actual risk of predation at natural nests for several reasons. First, all the studies use playback at artificial or translocated nests, where predation rates may differ from those at natural nests (Burke et al., 2004; Lindell et al., 2004; Major and Kendal, 1996; Pärt and Wretenberg, 2002; Thompson and Burhans, 2004; Weidinger, 2001). Second, begging calls were played back more continuously than occurs in natural nests. Third, some of the studies lacked replication of playback sounds (Kroodsma, 1990; Kroodsma et al., 2001). Finally, the absence of parents at experimental nests quite likely leads to an overestimate of the absolute costs of begging calls (Section III.A; Halupka, 1998; but see McDonald et al., 2009). Indeed, these experiments are perhaps most useful in showing what the cost of calling would be, were it not reduced by the many antipredator adaptations discussed throughout this review.