Results

One quarter of subjects tested in the presence of androstanone were unable to detect the heavily dosed androstanone strip (n=10): data from these subjects were discarded and replacement subjects were found. Two subjects in one of the no-audience conditions used the computer to help them produce unnaturally high scores in the trials, and their results were likewise discarded and replaced. None of the subjects when asked claimed to have detected any scent or other abnormality with any of the materials in the testing chamber. Most believed that the anosmic chamber was used because one could use it to isolate subjects from distracting stimuli.

Hypothesis One

Table 1 shows the mean, minimum and maximum number of characters typed (the score) by all subjects for both tasks and the standard deviation. Figure 1 shows a boxplot graph of the distribution of scores by task. The table and figure shows a much higher mean score for the easy task than it does for the hard task, so this suggests support for alternative hypothesis 1.

Table 1. Descriptive statistics for the Task Difficulty factor

	Easy Task	Hard Task
Mean	1021.11	164.95
Minimum	488	113
Maximum	2045	248
S.D.	328.87	32.34

Note: n=80. The difference between the means of the two levels of task difficulty was highly significant (p < 0.0005)

Figure 1.  Boxplot to show the distribution of scores for each level of task difficulty

Note the much higher scores for the easy task than the hard task. Note also the much wider distribution of scores in the easy task.  After taking the different sizes of the means into account, the S.D. for the easy task is still more than 50% bigger than that of the hard task.  This suggests it was not a consistently easy task for subjects to perform.

A 2 x 2 x 2 repeated measures ANOVA revealed a highly significant main effect of task difficulty on task performance F (1,76) = 659.76, p <0.0005. Hypothesis HA-1 is retained.

Hypothesis Two

Table 2 shows the means, minima, maxima and standard deviations for both the easy and hard tasks in the audience and no audience treatments. It can be seen from the means that, as predicted, in the audience condition subjects typed more characters in the easy task and fewer characters in the hard task than they did in the no audience condition. Figure 2 represents these means side by side. Note in this graph that in the audience treatment there is a noticeable trend for easy task scores to be higher and hard task scores to be lower than in the no audience treatment. The repeated measures ANOVA showed a significant interaction between audience treatment and task difficulty, F (1,76) = 11.37, p < 0.005. From this interaction and the direction of the difference in means (see Figure 2 and Table 2) hypothesis HA-2 can be retained.

Table 2. Descriptives for the Audience treatment and Task Difficulty.

		Audience	No Audience
Easy Task	Mean	1128.32	913.9
	Minimum	562	488
	Maximum	2045	1411
	S.D..	368.62	244.18
Hard Task	Mean	157.13	172.78
	Minimum	113	123
	Maximum	209	248
	S.D..	26.69	35.78

n=40. Note the differences between the means: in the audience treatment the subjects scored more highly on the easy task and less higly on the hard task than did subjects with no audience. This difference was significant (p < 0.005)
 
 

Figure 2.  Boxplots to show Audience effects on Easy and Hard tasks

Note in the audience treatment the scores are higher for the easy task and lower for the hard task than in the No audience treatment.  This is consistent with the predicted (and already well established in literature) effect of social faciliation.  The differences here are significant (p < 0.005)

Hypothesis Three

Table 3 shows the means, minima, maxima and standard deviations for each task difficulty across both levels of androstanone treatment. The subjects' mean score is higher for the easy task and lower for the hard task in the androstanone treatment than in the no androstanone treatment. Figure 3 represents this in a side by side boxplot of scores in the easy and hard tasks against the levels of androstanone treatment.

Table 3. Descriptives for the Androstanone treatment and Task Difficulty.

Note: The differences in the means and direction of these differences suggest that there may be an interaction between androstanone and task difficulty. However, in this experiment, no significant interaction was found (p > 0.05)
 

Figure 3.  Boxplots to show Androstanone effects on Easy and Hard tasks

Note in the androstanone treatment the scores are higher for the easy task and lower for the hard task than in the no androstanone treatment.  This is consistent with the predicted effect, but a repeated measures ANOVA found it to be non-significant (p > 0.05)

A repeated measures ANOVA did not confirm this relationship F (1,76) = 3.024, p > 0.05, so at this stage the third alternative hypothesis must be rejected, and null hypothesis H0-3 retained. However, given the relatively high S.D. (even accounting for the means being larger) of the easy task scores when compared with the hard task scores, it was thought that separate analysis of the effect of androstanone on each task's individual mean might reveal a significant effect. While analysis showed there was no significant effect of androstanone on scores in the easy task, a one-way ANOVA showed there was a highly significant effect of the androstanone treatment on scores in the hard task, F (1,78) = 23.953, p < 0.005. In the light of the significant effect of androstanone on the subjects performance in the hard task alone, hypothesis HA-3 can be partially retained. While subjects in the presence of androstanone did not type significantly more characters in the easy task than those who were not in the presence of androstanone, they did type significantly more characters in the hard task.

It is possible that a significant effect of androstanone presence on scores in the easy task may have been found had the sample size been larger. Maxwell and Delaney (1990) provide a method of determining the sample size required to show a significant effect of a certain power, given the predicted difference between means. Using power = 0.8 it is possible to use this method to determine what number of subjects this design would require, given the difference between the easy task means across both levels of the androstanone treatment. Where MA is the mean for the androstanone treatment and MNA is the mean with no androstanone, we can determine a value for d and look it up against the tables provided by Maxwell and Delaney:

The difference between the means in the easy task is 87.69. The S.D. is 328.87, so d is 0.26. The table shows that approximately 250 subjects would be required to achieve a significant effect (power = 0.8) given the small difference between means of the different treatments and the large variance between individuals' scores on the easy task.

When the subjects' sex was included in the statistical model (making a 2 x 2 x 2 x 2 within-between design) an interaction between subjects' sex and audience treatment emerged as significant F (1,72) = 5.047, p < 0.05. Examination of the means (see Table 4) reveals that men without an audience typed many more characters in the hard task than did women with or without an audience, or men with an audience. Table 4 also shows contrapositive findings: the mean score for women with an audience was higher in the easy task than for men with or without an audience, and women without one.

Table 4. Descriptives for the interaction Task Difficulty * Audience * Sex

			Mean	Min	Max	S.D.
Easy	Men	Audience	1019.6	725	1377	202.94
		No Audience	967.55	614	1344	216.83
	Women	Audience	1237.05	562	2045	461.35
		No Audience	860.25	488	1411	263.28
Hard	Men	Audience	155.86	113	209	28.41
		No Audience	183.75	123	248	44.47
	Women	Audience	158.39	115	209	25.53
		No Audience	161.8	133	201	19.91

Note: for each mean, n=20. The numbers in bold are those discussed below

This table shows the two treatment-task combinations with the very high means (men * no audience * hard task and women * audience * easy task) also have higher then normal standard deviations and maximum values, suggesting that outlying data may have lead to this interaction being significant. Indeed, there are a small number of outlying data points in all treatments, but, after checking for clerical errors, it was determined that these points were genuine data, and so were retained in the analysis. However, conducting the same ANOVA without the most extreme data points resulted in no significant interaction of task difficulty, audience and sex, F (1,66) = 3.1, p > 0.05, whereas all other interactions remained at similar p values.

Turn to the Discussion...