Electronics Repair Study

The baseline ORD repair has the mystery shopper state the required repair. The 'intervention' CRE repair has the mystery shopper state they have no idea what is wrong. The target of the intervention is the repair store, and the intention is to quantify any CRE markup in repair prices and compare these between countries.

2026-05-29

2026-11-13

The primary outcome is the credence markup, which we define as the final price paid for the CRE repair divided by the final price paid for the ORD repair times 100, calculated at the repair store level.

This experiment visits 30 different repair stores to conduct 60 repairs in 12 European countries. Each store conducts two repairs, a CRE(dence good) repair where the mystery shopper states they do not know what is wrong, and a ORD(inary good) repair where the mystery shopper states the fault and required repair. The list of possible repair stores are obtained from Google Maps, and this list is randomly shuffled with the stores at the top of the list being visited first until all observations are obtained. The order of the CRE or ORD visit is systematically assigned based on the order of this random list. The final price of the conducted repair is recorded, and the credence markup for the store is the CRE price divided by the ORD price. The data collection for this experiment will be conducted by a private survey agency.

Not available

The randomization method is the use of the random number generator present the pandas package of Python’s .sample() method, which shuffles the list of possible stores. Stores are then assigned based on the order
from the shuffled list, provided they are valid in that they currently repair the type of item in question

Repair store.

No

360 repair stores

12 countries × 30 markups = 360 markups (720 repair prices)

For each country, 30 repair stores will conduct 2 repairs each - a CRE repair and an ORD repair. The order of these visits is randomized based on the sequential order of the shuffled list. Some stores may only conduct 1 repair for whatever reason - these will be recorded but obviously cannot be used
to calculate a within-store credence markup. The stores that only conduct 1 repair will not count towards our target of 30 repair stores per country.

We conduct the following power analysis on our ability to detect a difference in credence markups between any two countries. We assume that prices within a country are generated by the following DGP: Price_{i,s} = β0 + β1CRE + v_{s} + ϵ_{i,s}, where v_{s} is a store-level random effect. To obtain values for the σ’s for v_{s} ∼ N (0, σ_{s}) and ϵ_{i,s} ∼ N (0, σ_{i}), we conduct a regression on data from a previous repair experiment on a different product that also conducted two visits per store. We multiply those previous prices by 2.5 to reflect the increased costs and cost range of the current type of repair, and obtain estimates of σ_{s} = 35 and σ_{s} = 27 and β0 = 115. Based on this transformed data, we also assume a minimum price of 75 (10% quantile). We set the maximum price to 350 Euros based on our experimental design. Given this DGP, we can generate prices for each country assuming different β1’s, calculate the credence markup (assuming CRE ≥ ORD), and estimate the regression: reg Markup Country, vce(robust). We assume Country 1 has β1 = 10, and simulate the smallest possible β1 for Country 2 so that we can detect the effect with 80% power at the 5% level. Using 1000 simulations, this value is β1 = 41.1, which implies an absolute difference in markups of 31.1 Euros, or for the relative difference in markups as defined of 21.7 units (e.g. we could detect the difference between a country with an average markup of 110 and another country with a markup of 131.7). This seems reasonable given we are mainly interested in substantial differences between countries. For context, Hall et al. (2019) report an equivalent markup of 144.4 in Turkey.