Advanced Orga 2

You dont have to pay taxes if you earn less than 450 euros. = mini jobbers

• Because of Aldi & Lidl --> tough time for this company
• Strategy change: If you change strategy you also have to think about organizational changes --> people need to become more service oriented
• Researches game and said: why dont you pay more to earn more? Tough discussion with management
• Before experiment no performance related salary.

• Why not individual: competition who serves which costumer. And in this store they could not track individual performance
• Why team incentives: Helping each other across shifts. Prepare for other shift ect.

If higher than target --> money 
if target reached --> 100 euros

Is divided among team on the basis of hours worked in this month.

Important note: Mini jobbers could not be included in the experiment. If they earn a bit more, they would need to pay taxes.

Distribution of sales in shops often not normal. To shift this graph into normal distribution we do a log of the y.
this is called log transformation.

• Randomization: put all the names of shops in a bowl and pick them out.
• Stratification = Block Randomization: Put shops into sublists (blocks) based on chosen characteristics. then you do the randomization whit each block. so kind of every block has the same amount of shops in each group. You increase the power of the design. For example in shops: You have really high sales in some control group and low sales in experimental treatment --> likely influences outcome of the experiment. If you make blocks of high income and low income shops, equally many shops with high performance will be in control and treatment group --> more comparable.
  --> especially important if characteristics strongly correlate with outcome variable
  --> can only be done with one or two variables
  • Two ways of stratification:
    1. Make several groups like on the picture
    2. Rank all the stores. assign 1&2 randomly to Control and Treatment. 3&4 randomly. 5&6 randomly ect.
• Stratification based on predicted parameters: regression analysis to be able to predict sales based on all possible parameters. according to prediction how good there sales will be, rank the shops. then do the same as with stratification before: put number 1&2 in a bowl and randomly chose, 3&4 ect.

stratification with prediction

beta coefficient gives an absolute increase

--> beta gives out a percentage, not absolute value

beta= 0.015 ---> increase 1.5%

We can use rule of thumb aslong as its below 0.1

they made a post regression: only data during treatment

and a change: where they included before and post (during) treatment data

Rule of thumb to calculate if significance
SD*2 < Coefficient --> significant 5%

SD*2= 0.034*2 = 0.068 --> not significant
SD*2= 0.013*2= 0.026 < 0.032 --> significant

--> they were not allowed to put the stars. in some papers they dont want, so readers calculate themselves and are aware of how much it is significant

it worked --> they implemented it for all stores.

--> no more effect as it was roled out across all stores

--> so no difference in the treatment for control and treatment groups

to answer how many participants for an experiment is needed.

-> we go for a power of 80% --> beta=0.2

Power = 1-beta

Power depends on cohens d, sample size and alpha

If cohens d higher --> power higher.  ---> its way easier to find a big effect than a small effect. need more sample siza for smaller effects.

If sample higher --> estimate becomes bigger --> sample distribution becomes thinner --> power gets bigger

and alpha. with bigger alpha --> bigger power

Cohen's d: how many standart deviations is H0 mean away from H1? (if they have same SD)

d=(Mean1-Mean2)/SD

rule of thumb: (can be really wrong, depending on area): 0.2= small effect. 0.5=medium effect. 0.8=large effect.

Small effect: n=395

Medium effect (d=0.5) --> n=63

Large effect (d=0.8) --> n=25

1. Identify Clusters: First, you identify clusters, which are groups of individuals that are naturally grouped together. These could be schools, villages, neighborhoods, or departments within a company.

2. Random Assignment: Instead of assigning treatments to individual people, you randomly assign the entire cluster to either the treatment group or the control group. For example, you might randomly choose which schools receive a new teaching method and which schools continue with the old method.

3. Apply the Intervention: Implement the intervention to the entire cluster. For instance, if a new teaching method is introduced, all teachers and students in the selected schools will follow the new method.

4. Measure Outcomes: Collect data from individuals within each cluster to measure the impact of the intervention. This could involve tests, surveys, or observations.

5. Analyze the Data: Compare the outcomes between the treatment clusters and the control clusters to determine the effectiveness of the intervention.

--> can lead that you need a much larger sample size!

--> can lead to that an experiment is not realisitic, becaus it would need so many people or clusters.

the smaller the cluster, the less bad the effect.

The experiment was conducted in 2 waves, why? 
--> 1st wave in May 2007: treatments Baseline, Money, Bottle and PriceTag. 
-->  2nd wave July/August 2010: treatments MoneyUpfront, Choice, Origami, additional Baseline treatment. 

Why was an additional Baseline treatment conducted in the 2nd wave? 
--> they wanted to see if the situation is still the same or if it changed over time. There was no difference in baseline treatment --> they pooled the data.

If you want to do additional treatments later you have to do something from the first wave treatment to make sure nothing changed.

Why: 
- Maybe they overestimated the price?

--> New treatment with saying the price and leaving a price tag: but not significantly lower than bottle treatment.

-Maybe they really liked the bottle?

--> new treatment "Choice" where they could test if they want the "Bottle Price Tag" or the money. 80% chose money. Still significantly better than baseline treatment.
--> So it was also not that they specifically like the bottle.

--> is it the time and effort invested by the employer? --> origami treatment: very large effect on productivity. 
--> it is the time and effort invested
 

Be aware: only for 3 hours. Not longterm tested. Also interesting if you stop giving the gift after some time.

- women earn less than men.
- big discussion in many countries
- aim to reduce the wage gap
- more transparency also for tax payers, so they know where there money goes.

Because this can lead to strong feelings of unfairness
can lead to big discussions

quasi experiment

gender gap --> reduced by 20%-40%

- less compensation
- more quitting
- public doesn't like high salaries in the public sector