• No results found

Experiments were only included if they assessed observer rats' tendency to respond in the same direction as their demonstrator after visual exposure to a demonstration session in which there was a contingency between demonstrated responses and an outcome. Docum entation for experiments for which this criteria was met, for at least some of the groups of observers, was retrieved from a set of records maintained at University College London which describes the bidirectional control experiments conducted by Heyes and associates. This set of records included all those for experiments conducted by the author during his PhD studentship, and all those that were available for experiments conducted by other investigators from the first bidirectional control experiment (Heyes & Dawson, 1990, conducted in December 1988) until the end of 1995^ Thus, the sample of experiments that were included covered the first seven years of research using the bidirectional control procedure, and the selection of these experiments was not biased to include only published findings. This avoids "the file drawer problem": When only published findings are retrieved, the average effect size is often an overestimation of the magnitude of an effect because publication biases favour statistically significant findings (see, e.g.. Smart, 1964; Smith,

1980).

Some experiments that were conducted in the surveyed time range could not be included. One experiment did not meet the criterion that observers were to be exposed to a demonstration session in which an outcome was contingent upon demonstrators responses (JS36, published as Heyes et al., 1993). In JS36, demonstrator-consistent responding was found to be inhibited when observed joystick pushes were followed by the omission of an appetitive reinforcer. Five further experiments could not be included because insufficient information was documented: In two of these cases (JS9a, IS 14) no data were present in the experimental record, and in each of the remaining three cases (JS9, JS34, JS38) there was

^The experiment reported by Reed et al. (1996) met these criteria, but was not included because it came to my attention after completing the meta-analysis.

Experimental records were not available for four study numbers (IS 15, JS30, JS31 & JS69), but it is unclear whether these studies were conducted.

not sufficient procedural information to interpret the data which were available. In addition, one experiment (JS72) was excluded because the investigator reported a serious bias in the equipment.

A total of 103 experiments remained. Several of these experiments examined demonstrator- consistent responding more than once. In some cases, the whole sample o f rats was subjected to more than one test. For example, in JS l (Heyes & Dawson, 1990) observers were subjected to a NDR test first, followed by a Reversal test, and then an Extinction test. When the animals' treatment in preceding tests was counterbalanced for each test, the results o f each test were treated here as being conceptually independent evaluations o f a demonstrator-consistent responding tendency. However, for the majority of cases in which demonstrator-consistent responding was examined more than once, the direction in which a demonstrator pushed the joystick was one factor in a between subjects factorial design; i.e., different animals were tested under different conditions. For the meta-analysis, each comparison was treated as being conceptually independent, although they occurred within the same experiment. Thus, an experiment such as JS32 (Heyes et al., 1992, Experiment 2) in which groups o f observers were exposed to either a left pushing or a right pushing demonstrator, but were tested either with the joystick in its original position (standard condition) or in a new position within the test chamber (joystick-transfer condition), contributed two estimations of the magnitude of a demonstrator-consistent responding effect, one for each type o f testing condition. Therefore, the units of analysis that were averaged for the estimation o f effect size were the evaluations of demonstrator-consistent responding that can be considered to be conceptually independent (see also Ucros, 1989, for the application of a similar approach). These units of analysis are described within this chapter as "study effects". A total of 165 study effects were yielded from the 103 experiments selected for consideration. Three of these study effects were excluded from the meta-analysis because they did not meet the minimal inclusion criteria: observers were not visually exposed to a demonstration session in which there was a contingency between demonstrated responses and an outcome^. Therefore, 162 study effects remained.

^The excluded study effects came from the following experiments: JS40 (reported as Heyes, Jaldow & Dawson, 1994; no outcome condition), JS65 (unpub.; non­

contingent food delivery during demonstration), and MG 18 (Experiment 4, this thesis; demonstrations in the dark condition)

Table 5.1: A summary of types of bidirectional control designs

Short code and Description

Group Names

Reinforced Direction Pre-

Treat Treat Test DV Prediction NDR

Non Differential LEFT L Both L/(L+R) LEFT

Reinforcement test RIGHT R Both > RIGHT

DR

Differential SAME L L D+/(L+R) SAME

Reinforcement test R R > D IF F

DIFF R L

L R

REV

Reversal test NEW L R R D-I-/(L+R) NEW

R L L > O L D

OLD L L R

R R L

EXT

Extinction test SAME L L 0 D+/(L+R) SAME

R R 0 > D IF F

DIFF L R 0

R L 0

TRAN

Transfer test following conditional discrimintation training. Procedures reported in full in Chapter 6.

offB

Shift from baseline LEFT Both L Both L/(L+R)pre LEFT RIGHT Both R Both -

L/(L+R)post

> RIGHT NoTrain_X

Extinction test LEFT L 0 L/(L+R) LEFT

without pretraining RIGHT R 0 > RIGHT

Notes. For each group tested under each type of design the following information is presented: The direction of joystick pushes for which an observer would be reinforced before (P re-treat) and after (Test) being exposed to a demonstration session; the demonstrated direction (Treat); the dependent variable (DV); and the predicted direction of a difference between group means expected for demonstrator-consistent responding.