|
Field
Trial End Date
|
Before
February 28, 2025
|
After
June 30, 2025
|
|
Field
Last Published
|
Before
December 20, 2024 01:43 PM
|
After
March 31, 2025 03:49 PM
|
|
Field
Intervention (Public)
|
Before
1. Trial: Participants are recruited as “managers” to assess the relative productivity of the older team member in "worker" teams of two who have to solve the Word Encryption Task.
There are three information treatments (see also Experimental Design below):
In NoInfo, the managers receive information about the joint distribution of the performance F(X), but not about the fact that the two workers come from different age groups.
In JointInfo, the managers receive information about the joint distribution of the performance F(X) and about the fact that the two workers come from different age groups.
In SepInfo, the managers receive information about the separate distributions of the performance F(X)_young and F(X)_old.
2. Trial
Participants are recruited as “workers” and have to solve the "Word Encryption Task" in a team, given information about the "managers" beliefs and different financial incentives.
There are five between-subjects treatments, a control treatment and 4 discrimination treatments that combine Bonus (fair, unfair) and discrimination based on performance information (AgeInfo, SepInfo). Note: Due to budget constraints and restrictions of the subject pool in the targeted older age group (52-59), the two Sep-Info treatments will only be implemented if the statistical power in terms of the minimum detectable effect size (MDE) in the AgeInfo treatments is sufficient.
Control: Each worker receives F’.
Fair-AgeInfo (FA): Each worker receives F’. The older worker is shown the managers’ expectations about the performance of the old workers from AgeInfo in the first Trial.
Fair-SepInfo (FS): Each worker receives F’. The older worker is shown the managers’ expectations about the performance of the old workers from SepInfo in the first Trial.
Unfair-AgeInfo (UA): The older worker is shown the managers’ expectations about the performance of the old workers from AgeInfo in the first Trial and receives only s_old|AgeInfo*2F’, whereas the younger worker receives (1-s_old|AgeInfo)*2F’.
Unfair-SepInfo (US): The older worker is shown the managers’ expectations about the performance of the old workers from SepInfo and the true s* in the first Trial and receives only s_old|SepInfo*2F’, whereas the younger worker receives (1-s_old|SepInfo)*2F’.
|
After
1. Trial: Participants are recruited as “managers” to assess the relative productivity of the older team member in "worker" teams of two who have to solve the Word Encryption Task.
There are four information treatments (see also Experimental Design below):
In NoInfo, the managers receive information about the joint distribution of the performance F(X), but not about the fact that the two workers come from different age groups.
In AgeInfo, the managers receive information about the joint distribution of the performance F(X) and about the fact that the two workers come from different age groups. The age group to which the two workers belong is not revealed.
In JointInfo, the managers receive information about the joint distribution of the performance F(X) and about the fact that the two workers come from different age groups. The age group to which the two workers belong is revealed.
In SepInfo, the managers receive information about the separate distributions of the performance F(X)_young and F(X)_old. The age group to which the two workers belong is revealed.
2. Trial
Participants are recruited as “workers” and have to solve the "Word Encryption Task" in a team, given information about the "managers' " beliefs and different financial incentives.
There are five between-subjects treatments, a control treatment and 4 discrimination treatments that combine Bonus (fair, unfair) and discrimination based on performance information (JointInfo, SepInfo). Note: Due to budget constraints and restrictions of the subject pool in the targeted older age group (52-59), the two Sep-Info treatments will only be implemented if the statistical power in terms of the minimum detectable effect size (MDE) in the JointInfo treatments is sufficient.
Control: Each worker receives F’.
Fair-JointInfo (FA): Each worker receives F’. The older worker is shown the managers’ expectations about the performance of the old workers from JointInfo in the first Trial.
Fair-SepInfo (FS): Each worker receives F’. The older worker is shown the managers’ expectations about the performance of the old workers from SepInfo in the first Trial.
UnfairJointInfo (UA): The older worker is shown the managers’ expectations about the performance of the old workers from JointInfo in the first Trial and receives only s_old|JointInfo*2F’, whereas the younger worker receives (1-s_old|JointInfo)*2F’.
Unfair-SepInfo (US): The older worker is shown the managers’ expectations about the performance of the old workers from SepInfo and the true s* in the first Trial and receives only s_old|SepInfo*2F’, whereas the younger worker receives (1-s_old|SepInfo)*2F’.
|
|
Field
Intervention End Date
|
Before
February 28, 2025
|
After
June 30, 2025
|
|
Field
Primary Outcomes (End Points)
|
Before
1. Trial:
The key outcome variable is the managers' belief about the performance of the older team member s_old.
We assume that the managers have distorted beliefs about the performance of the two age groups and therefore discriminate against older workers (even though this reduces their own payoff). In NoInfo, there can be no age discrimination because the age information is not available to the managers. Therefore, on average, the estimated proportion of s_old should be 50% (H1); in JointInfo, the managers have information that one of the two works is older, but they only know the joint distribution of the performance. So here, beliefs about F(X)_old and F(X)_young play a role and we hypothesize that on average s_old<50% is estimated (H2); finally, in SepInfo, false beliefs about age-related performance can be corrected. If s_old is still underestimated in relation to the old workers’ true average performance s*, it is pure taste-based discrimination, which is costly to the manager. Formally, we have
H1 (neutrality): s_old|NoInfo=0.5
H2 (belief-based discrimination): s_old|AgeInfo<0.5
H3 (taste-based discrimination): s_old|SepInfo<s*
With regard to the difference between AgeInfo and SepInfo, we expect that beliefs-based discrimination is stronger than taste-based discrimination.
H4 (belief- vs. taste-based discrimination): s_old|AgeInfo/0.5 < s_old|SepInfo/s*
At the within-subjects level, the observation over 10 rounds also tests whether the strength of discrimination or misbelief correlates with the level of overall performance, i.e., whether the older worker of stronger teams is systematically attributed an even smaller share of success. Formally, we expect
H5: d (s*-s_old)/dy>0.
Correlational Hypothesis:
We also hypothesize that bias against senior workers is more pronounced the younger the manager is. Formally, we expect that
H6: d(s*-s_old)/da<0, where a is the managers’ age.
2. Trial
The key outcome variable is the workers' performance in the Word Encryption Task s_old.
Hypotheses: We hypothesize that discrimination has a negative effect on the performance of older workers, with the discrimination effect being stronger for the unfair bonus and taste-based discrimination respectively.
H1 (bonus): s_old|C>s_old|U>s_old|F
H2 (info): s_old|C>s_old|A>s_old|S
H3 (interaction): s_old|FS<s_old|FA and s_old|US
|
After
1. Trial:
The key outcome variable is the managers' belief about the performance of the older team member s_old.
We assume that the managers have distorted beliefs about the performance of the two age groups and therefore discriminate against older workers (even though this reduces their own payoff). In NoInfo, there can be no age discrimination because the age information is not available to the managers. Therefore, on average, the estimated proportion of s_old should be 50% (H1); in AgeInfo, the managers have information that one of the two workers is older, but they do not know which. Therefore, on average, the estimated proportion of s_old should be 50% (H2); in JointInfo, the managers have information that one of the two works is older, but they only know the joint distribution of the performance. So here, beliefs about F(X)_old and F(X)_young play a role and we hypothesize that on average s_old<50% is estimated (H3); finally, in SepInfo, false beliefs about age-related performance can be corrected. If s_old is still underestimated in relation to the old workers’ true average performance s*, it is pure taste-based discrimination, which is costly to the manager. Formally, we have
H1 (neutrality): s_old|NoInfo=0.5
H2 (neutrality): s_old|AgeInfo=0.5
H3 (belief-based discrimination): s_old|JointInfo<0.5
H4 (taste-based discrimination): s_old|SepInfo<s*
With regard to the difference between JointInfo and SepInfo, we expect that beliefs-based discrimination is stronger than taste-based discrimination.
H5 (belief- vs. taste-based discrimination): s_old|JointInfo/0.5 < s_old|SepInfo/s*
At the within-subjects level, the observation over 10 rounds also tests whether the strength of discrimination or misbelief correlates with the level of overall performance, i.e., whether the older worker of stronger teams is systematically attributed an even smaller share of success. Formally, we expect
H6: d (s*-s_old)/dy>0.
Correlational Hypothesis:
We also hypothesize that bias against senior workers is more pronounced the younger the manager is. Formally, we expect that
H7: d(s*-s_old)/da<0, where a is the managers’ age.
2. Trial
The key outcome variable is the workers' performance in the Word Encryption Task s_old.
Hypotheses: We hypothesize that discrimination has a negative effect on the performance of older workers, with the discrimination effect being stronger for the unfair bonus and taste-based discrimination respectively.
H1 (bonus): s_old|C>s_old|U>s_old|F
H2 (info): s_old|C>s_old|A>s_old|S
H3 (interaction): s_old|FS<s_old|FA and s_old|US
|
|
Field
Experimental Design (Public)
|
Before
Creating the performance data set
We recruit participants in age group 22-29 (young) and participants in age group 52-59 (old) through Prolific UK. Each age group contains about 50% men and 50% women. The participants have to solve as many “Word Encryption Tasks” as possible within seven minutes. Both age groups receive the same piece rate W and a fixed fee F. Since the task involves letters, we recruit only participants who report not to have any literacy problems or language-related disorders.
Thus, we create two performance data sets F(X)_young and F(X)_old. Optional after visual inspection of the data: To minimize the effects of extreme outliers in relation to the distribution, we trim the upper and lower 5% of observations by replacing the respective data with the values of the 5th and 95th percentile.
1. Trial Phase:
Subjects (>35 years) are recruited at Prolific UK and assigned the role of “managers”. In 10 rounds, the managers are each given information about the overall performance y=x_young+x_old of a “worker” team of two workers, one randomly selected from F(X)_young and one from F(X)_old. In order to establish comparability between the treatments and subjects, 4 sequences of 10 teams are drawn at random before the start of the experiment and presented to a quarter of the managers in each case.
The managers' task is to estimate the older worker's relative contribution to the team’s overall performance: s_old=x_old/y.
There are three between-subjects treatments:
In NoInfo, the managers receive information about the joint distribution of the performance F(X), but not about the fact that the two workers come from different age groups.
In JointInfo, the managers receive information about the joint distribution of the performance F(X) and about the fact that the two workers come from different age groups.
In SepInfo, the managers receive information about the separate distributions of the performance F(X)_young and F(X)_old.
The more accurately the managers estimate the proportion s_old on average, the higher their payoff.
2. Trial Phase:
Young (aged 22-29) and old (planned: aged 52-59) people are recruited as “workers” through Prolific UK. They solve as many “Word Encryption Tasks” as they can and want in a timeframe of seven minutes. Both know that they work in a team that consists of one old and one young worker. Both age groups receive the same piece rate W and the team as a whole receives, depending on the treatment, a bonus corresponding to two times a fixed fee F’.
There are five between-subjects treatments, a control treatment and 4 discrimination treatments that combine Bonus (fair, unfair) and discrimination based on performance information (AgeInfo, SepInfo). Note: Due to budget constraints and restrictions of the subject pool in the targeted older age group (52-59), the two Sep-Info treatments will only be implemented if the statistical power in terms of the minimum detectable effect size (MDE) in the AgeInfo treatments is sufficient.
Control: Each worker receives F’.
Fair-AgeInfo (FA): Each worker receives F’. The older worker is shown the managers’ expectations about the performance of the old workers from AgeInfo in the first Trial.
Fair-SepInfo (FS): Each worker receives F’. The older worker is shown the managers’ expectations about the performance of the old workers from SepInfo in the first Trial.
Unfair-AgeInfo (UA): The older worker is shown the managers’ expectations about the performance of the old workers from AgeInfo in the first Trial and receives only s_old|AgeInfo*2F’, whereas the younger worker receives (1-s_old|AgeInfo)*2F’.
Unfair-SepInfo (US): The older worker is shown the managers’ expectations about the performance of the old workers from SepInfo and the true s* in the first Trial and receives only s_old|SepInfo*2F’, whereas the younger worker receives (1-s_old|SepInfo)*2F’.
|
After
Creating the performance data set
We recruit participants in age group 22-29 (young) and participants in age group 52-59 (old) through Prolific UK. Each age group contains about 50% men and 50% women. The participants have to solve as many “Word Encryption Tasks” as possible/as they choose to within seven minutes. Both age groups receive the same piece rate W and a fixed fee F. Since the task involves letters, we recruit only participants who report not to have any literacy problems or language-related disorders.
Thus, we create two performance data sets F(X)_young and F(X)_old. Optional after visual inspection of the data: To minimize the effects of extreme outliers in relation to the distribution, we trim the upper and lower 5% of observations by replacing the respective data with the values of the 5th and 95th percentile.
1. Trial Phase:
Subjects (>35 years) are recruited at Prolific UK and assigned the role of “managers”. In 10 rounds, the managers are each given information about the overall performance y=x_young+x_old of a “worker” team of two workers, one randomly selected from F(X)_young and one from F(X)_old. In order to establish comparability between the treatments and subjects, 4 sequences of 10 teams are drawn at random before the start of the experiment and presented to a quarter of the managers in each case.
The managers' task is to estimate the older worker's relative contribution to the team’s overall performance: s_old=x_old/y.
There are four between-subjects treatments:
In NoInfo, the managers receive information about the joint distribution of the performance F(X), but not about the fact that the two workers come from different age groups.
In AgeInfo, the managers receive information about the joint distribution of the performance F(X) and about the fact that the two workers come from different age groups. The age group to which the two workers belong is not revealed.
In JointInfo, the managers receive information about the joint distribution of the performance F(X) and about the fact that the two workers come from different age groups. The age group to which the two workers belong is revealed.
In SepInfo, the managers receive information about the separate distributions of the performance F(X)_young and F(X)_old.
The more accurately the managers estimate the proportion s_old on average, the higher their payoff. The age group to which the two workers belong is revealed.
2. Trial Phase:
Young (aged 22-29) and old (planned: aged 52-59) people are recruited as “workers” through Prolific UK. They solve as many “Word Encryption Tasks” as they can and want in a timeframe of seven minutes. Both know that they work in a team that consists of one old and one young worker. Both age groups receive the same piece rate W and the team as a whole receives, depending on the treatment, a bonus corresponding to two times a fixed fee F’.
There are five between-subjects treatments, a control treatment and 4 discrimination treatments that combine Bonus (fair, unfair) and discrimination based on performance information (JointInfo, SepInfo). Note: Due to budget constraints and restrictions of the subject pool in the targeted older age group (52-59), the two Sep-Info treatments will only be implemented if the statistical power in terms of the minimum detectable effect size (MDE) in the JointInfo treatments is sufficient.
Control: Each worker receives F’.
Fair-JointInfo (FA): Each worker receives F’. The older worker is shown the managers’ expectations about the performance of the old workers from AgeInfo in the first Trial.
Fair-SepInfo (FS): Each worker receives F’. The older worker is shown the managers’ expectations about the performance of the old workers from SepInfo in the first Trial.
Unfair-JointInfo (UA): The older worker is shown the managers’ expectations about the performance of the old workers from AgeInfo in the first Trial and receives only s_old|JointInfo*2F’, whereas the younger worker receives (1-s_old|JointInfo)*2F’.
Unfair-SepInfo (US): The older worker is shown the managers’ expectations about the performance of the old workers from SepInfo and the true s* in the first Trial and receives only s_old|SepInfo*2F’, whereas the younger worker receives (1-s_old|SepInfo)*2F’.
|
|
Field
Planned Number of Observations
|
Before
1200 participants (from different age groups, see below)
|
After
1400 participants (from different age groups, see below)
|
|
Field
Sample size (or number of clusters) by treatment arms
|
Before
For Trial Phase 1, we recruit 200 participants per treatment (600 in total). The sample will contain about 50% men and 50% women.
For Trial Phase 2, we recruit 120 participants per treatment (600 in total), of which 60 are from age group 22-29 (young) and 60 participants from age group old (planned: 52-59), respectively. The sample will contain about 50% men and 50% women. As noted above, depending on the MDE in the AgeInfo treatments, the number of participants in the AgeInfo treatments may be increased to 200 and the SepInfo treatments will not be carried out.
All samples will be recruited through Prolific UK.
|
After
For Trial Phase 1, we recruit 200 participants per treatment (800 in total). The sample will contain about 50% men and 50% women.
For Trial Phase 2, we recruit 120 participants per treatment (600 in total), of which 60 are from age group 22-29 (young) and 60 participants from age group old (planned: 52-59), respectively. The sample will contain about 50% men and 50% women. As noted above, depending on the MDE in the JointInfo treatments, the number of participants in the JointInfo treatments may be increased to 200 and the SepInfo treatments will not be carried out.
All samples will be recruited through Prolific UK.
|