I explored relationships between behavioural responses obtained in two affective state tests: the classical judgement bias test and an eyespot test in their original and here developed simplified versions.

In the original tests, the relationship between behavioural measurements from the two affective state tests persisted when removing effects from loss of ambiguity of the middle grey cue (linked to repeated exposures and learning). Individuals showing extreme aversion to eyespots also showed very long latencies to the middle grey cue. This confirms my original hypothesis that the relationship is present.

In the first array of the simplified tests, no relationship was detected. Individuals showing extreme aversion to eyespots were not the same who scored very long latencies to the middle grey cue. Moreover, removal of individuals showing extreme values in either test revealed a relationship, but only when removing effects of loss of ambiguity. This mismatch and dependence on extreme values indicates that individuals showing extreme values were responsible for disrupting the relationship, instead of leading it.

In the second array of the simplified tests, the relationship between behavioural measurements from the two affective state tests was disrupted by removing effects of loss of ambiguity. This indicates that birds learnt that the middle grey cue was unrewarded, and therefore that the judgement bias test could not properly detect affective states.

Overall, one of the used full eyespot cues presented the predicted correlation with responses to ambiguous cues from the judgement bias test. The full eyespot cue is therefore supported as aversive cue. However, it is still impossible to implement an eyespot test based on both full and ambiguous eyespot cues, as the latter showed no clear correlation with responses to ambiguous cues from the judgement bias test.

The original and simplified tests could detect the relationship between behavioural responses only in specific conditions, discussed below. The two versions of the tests differed for the experimental setup (e.g. number of test cues), age of test and number of tests carried out.

• Ability to detect the relationship with distance measurements: only the original tests were able to detect the relationship with the use of distance measurements. I suggest that the shorter runway length, used in the arena as adaptation for different ages, could have caused a ceiling effect when measuring distances from test cues in the simplified tests.
• Ability to detect the relationship with latency measurements: only the first array of simplified tests could not detect the relationship with the use of distance measurements. That array was carried out when birds were youngest. Age could affect how birds react to eyespot patterns, as chicks tested at older age with both the original and simplified tests showed fewer extreme values.
• Sex effects on detection of the relationship between tests: sex effects were detected only in the original test. Age could explain this result, as adult female fowls are known to be more motivated to work for food, and therefore testing chicks at older age might lead to more pronounced sex differences.
• Effects of extreme values on detection of the relationship between tests: when removing effects of loss of ambiguity, removal of outliers (detected only in the judgement bias test) made disappear all relationship between original tests. On the contrary, removal of outliers (detected mainly in the eyespot test) made the relationship appear between simplified tests in the first array under the same conditions of removal of loss of ambiguity effects. Thus, when accounting for loss of ambiguity, the original tests are more sensitive to extreme values from the judgement bias test, while the simplified tests are more sensitive to extreme values from the eyespot test. However, if not considering effects of loss of ambiguity, fewer extreme values were detected in both versions. Moreover, the original tests could still show the relationship, even with removal of extreme values. These effects indicate that, overall, results from both affective state tests from both versions are less affected by extreme values if not accounting for loss of ambiguity.