The effect of accent on wages and employment: Evidence from an experimental vignette study

The aim of the study is to examine whether a candidate’s spoken accent affects hiring decisions and wage offers in the United Kingdom using experimental vignettes. In particular, we study whether employers evaluate job applicants differently based on their regional accent and whether these effects differ across occupations with different communication demands.

Participants will be instructed to imagine themselves as hiring managers in a company recruiting for two positions: one sales position and one information technology (IT) position. They will evaluate hypothetical candidates applying for both jobs on the basis of candidate profiles (CVs) that include a name, education and work experience, and a short personal statement. In the main experimental condition, the personal statement will be presented as an audio recording reflecting different regional accents.

Accent is the primary treatment of interest and varies across four regional accent conditions that differ in perceived social status, including accents commonly perceived as relatively high status (for example, Edinburgh or Received Pronunciation) and accents commonly perceived as relatively low status (for example, Glaswegian or Newcastle). Job type distinguishes between an interaction-intensive occupation (sales) and a less interaction-intensive occupation (IT) and is used to assess whether accent effects differ across occupational contexts.

The primary outcomes are wage offers and hiring likelihood. Participants will report the hourly wage they would offer and the likelihood that they would hire each candidate using a 1-to-10 Likert-type scale.

We will test for a main effect of accent, expecting candidates with accents perceived as lower status to receive lower wage offers and lower hiring likelihood ratings than candidates with accents perceived as higher status.

We will also examine whether the effect of accent differs by job type, assessing whether accent penalties are more pronounced in interaction-intensive positions such as sales than in less interaction-intensive positions such as IT. These analyses are intended to evaluate heterogeneous treatment effects across occupational contexts, while the primary focus of the study remains the main effect of accent on hiring likelihood and wage offers.

Following the main experimental task, participants will answer a series of questions about the comprehensibility of the voices they previously heard and make social judgments about the speakers. In addition, participants’ demographic and background characteristics, including age, sex, education, and postcode information (first half only), will be recorded.

2026-06-01

2026-09-01

1. Wage Offer: Participants will be asked “What hourly wage would you offer this candidate?” and will respond using a continuous slider from 10 to 35 pounds.

2. Hiring Probability Decision: Participants will be asked “In a scale from 1 to 10, where 1 stands for “very unlikely” and 10 for “very likely”, how likely are you to hire this person?” and will respond using a 1-to-10 Likert-type scale.

1. Where do you think the speaker you just listened to is from? Please be as specific as you can. If you can identify the city they are from, please list that. If not, provide the region. (open ended)

2. How easy was it for you to recognize the speaker’s accent?
7-point Likert scale (1 = Very difficult, 7 = Very easy)

3. How confident are you in your guess about where the speaker is from?
7-point Likert scale (1 = Not at all confident, 7 = Very confident)

4. Easy to understand
7-point Likert scale (1 = Not at all, 7 = Very much)

5. To what extent was the speaker pleasant
7-point Likert scale (1 = Not at all, 7 = Very pleasant)

Social judgement questions: To what extend do you agree with the following statements about the applicant?

1. He was intelligent
7-point Likert scale (1 = Strongly Disagree, 7 = Strongly Agree)

2. He was successful
7-point Likert scale (1 = Strongly Disagree, 7 = Strongly Agree)

3. He was competent
7-point Likert scale (1 = Strongly Disagree, 7 = Strongly Agree)

4. He was trustworthy
7-point Likert scale (1 = Strongly Disagree, 7 = Strongly Agree)

5. He was pleasant
7-point Likert scale (1 = Strongly Disagree, 7 = Strongly Agree)

6. He was friendly
7-point Likert scale (1 = Strongly Disagree, 7 = Strongly Agree)

7. He would be productive in his job
7-point Likert scale (1 = Strongly Disagree, 7 = Strongly Agree)

Please complete the following questions about your background and education.

1. Age: Slider

2. Sex: ☐ Male ☐ Female ☐ Other ☐ Prefer not to say

3. What is your highest level of qualification?

☐ 1: High (Tertiary Education), i.e. Degree or equivalent and above

☐ 2: Medium (Upper Secondary or Further Education), i.e. post-compulsory but below degree level (e.g. A-Levels, Higher and Advanced Highers in Scotland, BTEC National Diploma)

☐ 3: Low (No Qualification to Lower Secondary Education), e.g. GCSEs, National 5 (N5) in Scotland

4. Is English your first language? ☐ Yes ☐ No

5. Was English the main language spoken at home when you were growing up?
☐ Yes ☐ No

5. First part (three digits) of the postcode where you grew up (If you moved during childhood, please provide the first part of the postcode (three digits) you most associate with growing up):

6. First part (three digits) of the postcode where you currently live:

The study will recruit participants through Prolific (prolific.com) and offer up to £3.50 for participation, as the expected completion time is approximately 10 minutes. Eligible participants must be adult native English speakers currently residing in the United Kingdom and have normal or corrected-to-normal hearing. Individuals with uncorrected hearing difficulties will be excluded.

The study employs a within-subjects online survey experiment designed to estimate the causal effect of accent on hiring decisions and wage offers across occupational contexts. Accent is the primary treatment of interest and has four levels, while job type provides the second treatment dimension with two levels: sales and information technology (IT). Each respondent evaluates four candidate profiles in total: two candidates for a sales vacancy and two candidates for an IT vacancy. This structure allows the study to examine whether the effect of accent differs across occupations that vary in the importance of verbal interaction.

Four profile attributes are experimentally manipulated: name (four levels), accent (four levels), personal statement (two levels within sales and two levels within IT), and education and work experience (two levels within sales and two levels within IT). Name and accent vary across all profiles, whereas personal statement and education/work experience are specific to job type. Name, personal statement, and education/work experience serve as control attributes designed to maintain realism and comparability across candidate profiles, whereas accent defines the main experimental contrast of substantive interest and job type is used to assess heterogeneous effects across occupational contexts.

Randomization is implemented at the individual respondent level using a balanced within-subject profile design. Each respondent is assigned a structured set of four candidate profiles such that all four accent conditions appear exactly once and all four name conditions appear exactly once. Within each job type, the two levels of personal statement and the two levels of education and work experience are balanced across the two candidate profiles. This construction ensures that comparisons across accents are made within respondent while maintaining balance across the main experimental attributes.

Presentation order is counterbalanced separately using a Latin square design. After the four candidate profiles are assigned, the Latin square rotates the sequence in which profiles are presented across respondents so that each profile type appears in different presentation positions with equal frequency. This procedure reduces potential order and sequence effects while preserving balance in exposure. Thus, profile attributes are balanced by construction within respondent, while the Latin square is used specifically to counterbalance presentation order across respondents.

The experiment will include four accents selected from the pilot study to represent differences in perceived social status: two relatively high-status accents (one English and one Scottish) and two relatively low-status accents (one English and one Scottish). Example accents may include Received Pronunciation (RP) or Estuary English, Edinburgh, Glaswegian, and Newcastle accents. Four speakers will be chosen for the main experiment, one representing each accent condition. In addition, 50 participants will be assigned to a control condition in which the candidate’s personal statement is presented in written rather than audio form. This control condition is intended to provide descriptive evidence on whether candidate profiles are perceived as similarly suitable in the absence of spoken accent information.

The primary dependent variables are as follows. First, participants will be asked, “What hourly wage would you offer this candidate?” and will respond using a continuous slider from 10 to 35 pounds. Second, participants will answer: “In a scale from 1 to 10, where 1 stands for “very unlikely” and 10 for “very likely”, how likely are you to hire this person?” Responses will be recorded using a 1-to-10 Likert-type scale.

The candidate profiles (CVs) and job descriptions will be created specifically for this research, drawing on real-world examples from job postings and publicly available CV formats, such as those found on LinkedIn, while ensuring that all candidate details—including employer names, addresses, work histories, and listed skills—are entirely fictional. To ensure balance across conditions, the CVs will be constructed to be as similar as possible in overall quality and experience. For example, each candidate will have two previous job experiences: one emphasizing stronger communication skills and one more closely tailored to the specific job post, both covering comparable time periods. Information from a previous employer indicating strong past performance will also be included and fixed for all candidates. This is intended to hold perceived productivity constant across candidates and reduce confounding between productivity and the experimental manipulations.

The study will be conducted online using the GORILLA experiment platform. Prior to the start of the experiment, informed consent will be obtained. Before the main task begins, participants will complete a brief practice round with an example profile and audio recording to familiarize themselves with the procedure.

Participants will then complete the experimental task by reviewing each job description and evaluating the corresponding candidate profiles.

After the experimental task is completed, participants will answer a series of questions about the comprehensibility of the voices they previously heard, as well as make social judgments about the speakers. Participants assigned to the written-statement control condition will not complete this section, since they will not have heard the audio stimuli.

Finally, participants will complete a short demographic and background questionnaire, including questions on age, sex, education, language background, and geographic background, including current residence and birthplace postcode (first half only). Participants will then be thanked, debriefed, and compensated.

Not available

Randomization will be done by a computer using Gorilla software

Randomization is implemented at the individual respondent level. Each respondent evaluates four candidate profiles in total: two candidates for a sales position and two candidates for an information technology (IT) position. Accent and name are assigned across profiles so that each respondent is exposed to all four accent conditions and all four name conditions exactly once across the four profiles. Within each job type, the two levels of personal statement and the two levels of education and work experience are balanced across the two candidate profiles. This balanced within-subject design ensures that the main effect of accent is identified through within-respondent comparisons while maintaining comparability across candidate profiles.

To reduce potential order and sequence effects, the order of profile presentation is counterbalanced across respondents using a Latin square design. This ensures that profile types appear in different presentation positions with equal frequency while preserving balanced exposure to the experimental conditions. The within-subjects structure improves statistical efficiency and facilitates within-respondent comparisons of the main effect of accent, while also allowing the study to examine whether accent effects differ across job types.

No

0

950 individuals

950 individuals

Power calculations were conducted by simulation to match the planned experimental design and estimation strategy. The simulation reproduced the within-subjects vignette structure in which each respondent evaluates four candidate profiles (two sales and two IT), with balanced assignment of accent and control attributes across profiles and counterbalancing of presentation order using a Latin square. The planned analysis uses linear regression with respondent fixed effects and standard errors clustered at the respondent level. For the continuous main outcome (log wage offer), using a significance level of 0.05 and target power of 0.80, the minimum detectable effect size (MDES) for a single focal treatment coefficient was estimated at approximately 0.04 log points, corresponding to about a 4 percent difference in the outcome. This represents the smallest effect that the study is expected to detect with 80 percent power under the assumed simulation parameters. Power calculations for the interaction between accent and job type were also conducted using the same simulation framework. For a single interaction coefficient between accent and job type in the continuous-response model, the MDES was estimated at approximately 0.09 log points, corresponding to about a 9 percent difference in the outcome. The study is primarily powered to detect main effects of accent on wage offers and hiring likelihood, while analyses of heterogeneity across job types are designed to detect only moderate-to-large differences in accent effects across occupational contexts. Accordingly, the primary focus of the study is the estimation of average accent effects, with job type used to assess whether these effects vary meaningfully across occupations with different communication demands. Both MDES estimates refer to single pre-specified treatment contrasts rather than to joint tests across multiple coefficients.