A field experiment on ethnic bias in public housing practices in Sweden

Figures & data

Figure 1. Letter of inquiry.

Note: English translation of the two original letters of inquiry in Swedish sent from the fictitious couples to the public housing companies. The first version of the letter was used for all inquiries on January 18, 2022, and the second version was used for all inquiries on April 2, 2022. The names of the couples (marked red in the figure) were randomly assigned in the first stage of mailing and alternated in the second stage. In the second version of the letter, the actual name of a municipality (marked green in the figure) was given.

Table 1. Definitions of variables.

Variables	Explanation
Outcome variables
Email reply	1 if a public housing company responded to a fictitious couple with a nonautomated letter; 0 otherwise.
Greeting	1 if a public housing company initiated its response to the couple with a personal salutation; 0 otherwise.
Cut in line	1 if a public housing company answered that they had a priority system for newcomers; 0 otherwise.
Bad zones	1 if the public housing company informed the couple about problematic neighborhoods; 0 otherwise.
Explanatory variable
Arab	1 if the inquiring couple had typical Arabic-sounding names; 0 if the couple had typical Swedish-sounding names.
Control variables
Large county	1 if the letter of inquiry was sent to a public housing company in one of the three largest counties in Sweden; 0 otherwise.
First batch	1 if the letter of inquiry belonged to the first batch of emails sent; 0 if it belonged to the second batch of email sent.

Note: The first batch of emails were sent January 18, 2022, and the second batch April 2, 2022. The three largest counties in Sweden are: Skåne län, Västra Götalands län, and Stockholms län.

Figure 2. Fraction of inquiries that led to a positive outcome for each couple.

Note: Blue lines are 95% CI. № of firms = 271. Each firm received two inquiries, one from each couple. Red dots represent the probability of receiving an email reply (A), personal greeting (B), information about priority system for newcomers (C), and information about bad neighborhoods (D). Definitions of all variables are provided in Table 1.

Table 2. Distribution of public housing companies’ responses to housing inquirers.

Outcome	No of firms	Neither	Both	Swedish	Arab	Difference
Email reply	271	26	168	38	39	1
Greeting	271	187	23	38	23	–15*
Cut in line	271	155	30	41	45	4
Bad zones	271	240	4	23	4	–19***

Note: Neither and Both are the number of cases in which both couples failed and succeeded in their correspondence with the public housing company, respectively. Swedish and Arab are the number of cases in which only the Swedish and Arab couple, respectively, succeeded to achieve a positive outcome in correspondence with the public housing companies. Difference is net differential treatment, i.e. the difference between cases given in columns Arab and Swedish. The null hypothesis of symmetry between the number of public housing companies who favored the Swedish couple and the number of public housing companies who favored the Arab couple was evaluated using the McNemar’s test and a two-sided binomial exact test. Definitions of all variables are provided in Table 1.

*** p < 0.01, **p < 0.05, *p < 0.10.

Table 3. Positive outcome probabilities, adjusted for county and order effects.

	Model 1 Email reply	Model 2 Greeting	Model 3 Cut in line	Model 4 Bad zones
Constant	0.692*** (0.034)	0.168*** (0.030)	0.287*** (0.034)	0.100*** (0.021)
Swedish	Reference	Reference	Reference	Reference
Arab	–0.001 (0.032)	–0.058** (0.028)	0.011 (0.034)	–0.070*** (0.019)
Large county	0.044 (0.041)	0.060 (0.042)	–0.173*** (0.040)	0.009 (0.025)
First batch	0.107*** (0.032)	0.072** (0.028)	0.081** (0.034)	–0.008 (0.019)
N	542	542	542	542
R^2	0.018	0.018	0.044	0.021

Note: The table presents positive outcome (or public housing companies’ response) probabilities (for our four outcomes) as a function of the inquiring couple’s ethnicity, estimated using linear probability models. Definitions of all variables are provided in Table 1. Reported standard errors (in parentheses) are corrected for clustering of the observations at the public housing company level. Corresponding probit models yield similar results.

*** p < 0.01, **p < 0.05, *p < 0.10.