A living wage vs UBI: Experimental evidence of wage floor framing and anchoring effects on redistributive preferences

We will recruit 2,000 from the general adult US population in using the platform Prolific. Participants will be redirected to the Qualtrics platform to begin the survey. During the survey participants will be redirected twice to another survey platform, qvsr.io, to complete a Quadratic Voting (QV) ‘game.’

The framing and anchoring experimental treatments are a Living Wage and a $18/hour state wage respectively (see Table 1). Experimental treatments will not be mutually exclusive. The four reference and treatment groups are seen in Table 1 & Figure 1, with vignettes in Table 2 and Figure 2. Participants will be evenly distributed to the four groups.

2024-08-28

2024-08-31

H1a. A living wage frame will increase support for a state-level regulatory wage floor.
H1b. A living wage frame will increase the preferred monetary wage floor preferences, both at the national and state levels.


H2a. A $18/hour policy will decrease support for a state-level regulatory wage floor.
H2b. A $18/hour policy will increase agreement that the wage floor will lead to unemployment and hurt the economy.
H2c. A $18/hour policy will have positive effects on the preferred monetary wage floor preferences, both at the national and state levels.

We posit that the politically salient term living wage will have noticeable effects on public support and attitudes, even when controlling for the monetary amounts of the wage. The living wage frame implies a higher wage floor than a minimum wage, and therefore we hypothesize that using this frame will nudge the general population towards preferring a higher preferred hourly wage for those at the bottom on the income ladder, as previously demonstrated in the UK (Schaitberger, 2024). In other words, we hypothesize participants will react differently to a policy of a $11 living wage compared to a policy of a $11 minimum wage, with similar differences between a $18 minimum and living wage, and this will have quantifiable effects on their preferences and attitudes.

We expect a high proposed increase to the wage floor could decrease public support, even in the face of broad public support, along with increasing participant beliefs that the policy will have negative economic effects. This result would be similar to those found on UK participants (Schaitberger, 2024). However, we also argue a policy of a large monetary increase will act as a numeric anchor, raising the preferred levels of both state and national wage floors (Brewer and Chapman, 2002; Buncic et al., 2021). While these hypotheses may seem intuitive to any observer, they have yet to be empirically tested on American participants in the academic literature (Searle, 2020).

H1c. A living wage frame will have positive or null effects on support for other redistributive policies, specifically UBI.
H2d. A $18/hour policy will decrease UBI support and preferred UBI monthly payments.
H2e. A $18/hour policy will decrease support for other redistributive policies.

The frame living wage is more benevolent and ambiguous (Bennett, 2012; 2014) while the higher proposed wage is more salient with the numeric anchor (Arceneaux and Nicholson, 2024). The frame depicts a general system, while the wage can be interpreted as a direct pay increase for the lowest paid worker. For these reasons, we submit H1 with positive redistributive effects while H2 with negative spillover effects on economic redistribution to those at the bottom of the income ladder.

3. Experimental Design
Our research design employs a choice experiment between a current and alternative public policy, with a frame of the alternative policy acting as the treatment (Elias et al., 2019; Lennon et al., 2022). We diverge from the work of Lennon et al. by testing wage floor preferences, expectations, and support for other redistributive policies, without any prompting of unemployment effects. Additionally, to test the framing effect of the wage name, we manipulate which frame is received by participants. The full survey instrument can be found in the Appendix.
3.1 Sample and Implementation
American citizens aged 18+ will be the study population. We will recruit 2,000 from the general adult US population in using the platform Prolific. Participants will be redirected to the Qualtrics platform to begin the survey. During the survey participants will be redirected twice to another survey platform, qvsr.io, to complete a Quadratic Voting (QV) ‘game.’ Qualtrics employs a function to divide participants to different treatment groups with an equal probability, ensuring fair and equal randomization. An example of the full survey can be found here:
https://kclbs.eu.qualtrics.com/jfe/form/SV_b2B6Ms52zz1GVue.

3.2 Baseline QV
Given the unique nature of a Quadratic Voting Survey, a baseline QV game was used to familiarize participants with the research tool and allow for collecting baseline preferences of participants prior to administering treatments (Cavaillé, Chen & Van Der Straeten, 2019). Participants were shown a brief 90 second video on QV: https://youtu.be/GrY_RzDsqLY. Then a link to a QV game will be provided, where participants will have 25 points to assign to 5 different government policies that they either support or disapprove of each policy. However, QV requires all additional points to be counted quadratically, meaning points could be assigned as 1 vote = 1 point, 2 votes = 4 points, 3 votes = 9 points, 4 votes= 16 points, 5 votes = 25 points. The operationalization of QV preference will be an 11-point scale with -5 indicating five votes against the policy, 0 indicating no votes on the policy, and 5 indicating five votes in favor of the policy Therefore, a strong preference would cost substantially more points than a weak preference, and the strongest preference of 5 would use all points to just a single policy. Once the Baseline QV was completed, participants are redirected back to Qualtrics for the experimental treatment vignette. The baseline QV can be found at: https://qvsr.io/survey/Zi0w4ogUAYOCLUMEKprN.

3.3 Experimental Treatments
This experiment uses a factorial, cross-cutting design, which are “widely used to study multiple treatments in one experiment” (Muralidharan et al., 2023: 1). The framing and anchoring experimental treatments are a Living Wage and a $18/hour state wage respectively (see Table 1). Experimental treatments will not be mutually exclusive. The four reference and treatment groups are seen in Table 1 & Figure 1, with vignettes in Table 2 and Figure 2. Participants will be evenly distributed to the four groups.

There will be one reference group and three treatment groups. Therefore, the state level alternative wage floor policies are as follows: (1) a $11/hour state minimum wage; (2) a $11/hour state living wage; (3) a $18/hour state minimum wage; and (4) a $18/hour state living wage. The first group will act as a reference group, while the remaining three groups will act as independent treatment groups which are prescribed a treatment frame, anchor or both. Participants will be asked to compare the existing and prescribed alternative policy, answer preference questions.

Overall, this choice experiment design follows the methods used by Elias et al. (2019) and Lennon et al. (2022), through directly asking participants to compare a current and a proposed policy. Our dependent variable, participant preferences for minimum wage and economic redistribution, is rooted this decision analysis between a current policy and a policy proposal. After the choice is presented, the preferences for the competing wage systems will be recorded via binary questions of support for the current policy or proposed policy. This includes asking participants to compare the current federal minimum wage policy to one of the alternative policies regarding support for a state mandated wage floor, expected unemployment effects, and desire for cities to also have their own wage floors. These questions are followed by asking participants to choose their preferred federal and state wage floor. These first five questions permit participants to directly compare both policies, and then we will ask participants to focus on the alternative policy received after reflecting on how it compares to the existing policy.

Next, treatment groups will be asked to imagine one of the alternative policies is enacted (see Figure 1), with the reference group prescribed a regulatory state wage that matches the norm as of 2024 (Department of Labor, 2024). After this policy prescription, participants will be asked additional questions related to economic redistribution and UBI. An attention check question is also asked to ensure compliance and reinforce the monetary figures within the experiment. Once participants have answered questions in Qualtrics, they will be redirected to complete another QV. However, this time participants will be instructed to answer as if the prescribed alternative wage floor system was implemented. Overall, the sliders, numeric-textbox, Likert scale questions and QV should provide robust evidence of the framing and anchoring effects from both the term living wage and a high proposed regulatory wage floor. Please see the following link for an example post treatment QV prescribed to participants: https://qvsr.io/survey/8cCD4ByX07jZ8eWA7p6z.

Our focus is on estimating the effect of our treatments on support wage floor, redistribution and UBI. Our empirical strategy employs a cross-cutting, also called a factorial design, that accounts for all three treatments for testing each of these hypotheses, along with controlling for any interaction (Kremer, 2003). We will use OLS to estimate effects to estimate effects using the following equations for the full sample:

(See Attached)

For all equations, variables used for regression analysis include the dependent variable Preference for the individual i. Preference will be measured using binary choice questions, Likert scale questions measuring preferences on a 7-point scale, a monetary slider, and a 11-point QV method. Regarding treatment variables, Living, $11, and $18 are dummy variables indicating treatment status, with β1, β2, β3 denoting the coefficients resulting from the participant exposure to the respective treatments. β0 is the intercept, and εi is the error term. Covariates in each equation are denoted with Z for the coefficients for up to K covariates, which include gender, age, ethnicity, income, education, partisanship, political ideology, and employment status. Additionally, we will test heterogenous treatment effects by analyzing whether effects vary by major socio-economic characteristics, such as age, race, and partisanship.

Qualtrics employs a function to divide participants to different treatment groups with an equal probability, ensuring fair and equal randomization. An example of the full survey can be found here in the attached Pre-Analysis Plan.

frames: Living Wage or Minimum Wage
anchors: $11/hour or $18/hour

No

We will recruit 2,000 from the general adult US population in using the platform Prolific.

2,000 observations, with one trial per participant.

500 Reference ($11/hour minimum wage), 500 Treatment 1 ($11/hour living wage), 500 Treatment 2, ($18/hour minimum wage), 500 Treatment 3 ($18/hour living wage)

A power study of the results for effects on wage floor support along with spillover effects on UBI and other redistributive support was conducted. A power estimate of 0.8 and an alpha of 0.05 was used with the coefficient and standard deviation for each dependent variable of interest, with the STAT 18.0 power analysis tool to produce a total N required based on the pilot study data. The N total observations required for a MDE to either reject or accept the null hypothesis, based on the pilot study data, for each variable of interest is seen in the right most column of Table 3. Surprisingly, the pilot study met some standards for an MDE. However, all variables listed in Table 3 are of interest to the final project. An MDE for all 9 questions is 2,598 for at least one of the treatments or QV reference per dependent variable. If we exclude the QV reference group, that number drops to 1,206 for one of the two treatments likely to have statistically significant effects on each of the 8 questions gauged. Considering the MDE’s based on the pilot data, we propose a sample of 2,000 from the general population. This will meet the MDE for each variable of interest.

A living wage vs. Universal Basic Income