The giving advice effect: Reducing teacher sorting through self-persuasion

Ajzenman, Nicolas; Elacqua, Gregory; Kutscher, Macarena; Méndez, Carolina; Suáre, Sonia Working Paper The giving advice effect: Reducing teacher sorting through self-persuasion IDB Working Paper Series, No. IDB-WP-01689 Provided in Cooperation with: Inter-American Development Bank (IDB), Washington, DC Suggested Citation: Ajzenman, Nicolas; Elacqua, Gregory; Kutscher, Macarena; Méndez, Carolina; Suáre, Sonia (2025) : The giving advice effect: Reducing teacher sorting through self-persuasion, IDB Working Paper Series, No. IDB-WP-01689, Inter-American Development Bank (IDB), Washington, DC, https://doi.org/10.18235/0013442 This Version is available at: https://hdl.handle.net/10419/315945 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/3.0/igo/ The Giving Advice Effect: Reducing Teacher Sorting Through Self-Persuasion Nicolas Ajzenman Gregory Elacqua Macarena Kutscher Carolina Méndez Sonia Suárez‖ WORKING PAPER No IDB-WP-01689 Inter-American Development Bank Education Division February 2025 The Giving Advice Effect: Reducing Teacher Sorting Through Self-Persuasion Nicolas Ajzenman Gregory Elacqua Macarena Kutscher Carolina Méndez Sonia Suárez Inter-American Development Bank Education Division February 2025 Cataloging-in-Publication data provided by the Inter-American Development Bank Felipe Herrera Library The Giving Advice Effect: Reducing Teacher Sorting Through SelfPersuasion / Nicolas Ajzenman, Gregory Elacqua, Macarena Kutscher, Carolina Mendez, Sonia Suarez. p. cm. — (IDB Working Paper Series; 1689) Includes bibliographical references. 1. Teacher transfer-Peru. 2. Teachers-Selection and appointment-Peru. 3. Teachers-Supply and demand-Peru. 4. Economics-Psychological aspectsPeru. I. Ajzenman, Nicolás. II. Elacqua, Gregory M., 1972III. Kutscher, Macarena. IV. Méndez, Carolina. V. Suarez Enciso, Sonia. VI. Inter-American Development Bank. Education Division. VII. Series. IDB-WP-1689 Jel Codes: D91,I23, I25 Keywords: Teachers, teacher policy, teacher shortages, behavioral, giving advice http://www.iadb.org Copyright © 2025 Inter-American Development Bank ("IDB"). This work is subject to a Creative Commons license CC BY 3.0 IGO (https://creativecommons.org/licenses/by/3.0/igo/legalcode). The terms and conditions indicated in the URL link must be met and the respective recognition must be granted to the IDB. Further to section 8 of the above license, any mediation relating to disputes arising under such license shall be conducted in accordance with the WIPO Mediation Rules. Any dispute related to the use of the works of the IDB that cannot be settled amicably shall be submitted to arbitration pursuant to the United Nations Commission on International Trade Law (UNCITRAL) rules. The use of the IDB's name for any purpose other than for attribution, and the use of IDB's logo shall be subject to a separate written license agreement between the IDB and the user and is not authorized as part of this license. Note that the URL link includes terms and conditions that are an integral part of this license. The opinions expressed in this work are those of the authors and do not necessarily reflect the views of the Inter-American Development Bank, its Board of Directors, or the countries they represent. The Giving Advice Effect: Reducing Teacher Sorting Through Self-Persuasion∗ Nicolas Ajzenman†Gregory Elacqua‡Macarena Kutscher§ Carolina M´endez Vargas¶Sonia Su´arez‖ February 13, 2025 Abstract This paper examines how the act of giving advice to others can serve as a tool for self-persuasion in high-stakes decisions. We tested this hypothesis in Peru’s nationwide teacher selection process, involving over 74,000 candidates. By prompting teachers to advise peers on selecting schools for maximum educational impact, we observe a significant shift in their own choices: an increased probability of choosing and being assigned to hard-to-staff schools, institutions serving disadvantaged areas that are typically understaffed. In line with recent literature on behavioral sciences, our findings demonstrate that advising others can influence one’s own consequential decisions. This insight offers a cost-effective approach to mitigating teacher sorting and reducing educational inequality. It also corroborates the validity of the giving advice effect in a high-stakes, real-world context using a large sample. JEL classification: D91,I23, I25 Keywords: Teachers, teacher policy, teacher shortages, behavioral, giving advice ∗Trial registered with the number AEARCTR-0012373. Research reviewed by the McGill University Research Ethics Board. We gratefully acknowledge the Inter-American Development Bank for funding the research presented herein. The opinions expressed in this publication are those of the authors and do not necessarily reflect the views of the Inter-American Development Bank, its Board of Directors, or the countries they represent. The authors have no conflicts of interest or financial and material interests in the results. All errors are our own. †McGill and IZA. E-mail: nicolas.a[email protected] ‡Inter-American Development Bank. E-mail: gregory[email protected] §Inter-American Development Bank. E-mail: [email protected] ¶Inter-American Development Bank. E-mail: [email protected] ‖Inter-American Development Bank. E-mail: [email protected] 1 1 Introduction A fundamental principle of public education systems worldwide is ensuring equal opportunities for students from diverse socioeconomic backgrounds. However, many countries face a pervasive challenge that threatens this objective: the phenomenon known as teacher sorting (Jackson,2009;Lankford et al.,2002;Boyd et al.,2013;Pop-Eleches and Urquiola, 2013). This issue arises when students from low-income households are more likely to attend understaffed schools with less qualified teachers. Teachers play a crucial role in the education process (Rivkin et al.,2005); their influence is typically greater for students from hard-to-staff backgrounds (Araujo et al.,2016). The shortage of high-quality educators in more vulnerable areas has severe detrimental implications for equity, as teacher sorting tends to exacerbate potential achievement gaps (Sass et al.,2012;Thiemann,2018). With a few recent exceptions (Ajzenman et al.,2024a,b), policy responses have predominantly centered on increasing compensation for positions at understaffed schools to motivate teachers to include such hard-to-staff schools among their potential choices (Evans and Mendez Acosta,2023). Although such measures may influence applicants’ decisions in specific cases (Neilson et al.,2021;Pugatch and Schroeder,2014), they often prove ineffective (Glazerman et al.,2012;Rosa,2017) and are generally associated with substantial costs. In this paper, we present the results of an experimental assessment of a nationwide, real-world, zero-cost nudge to mitigate teacher sorting in the Peruvian national public school teacher selection process in 2022 (CPM 2022, for its acronym in Spanish), inspired by a recent insight from the behavioral sciences literature: the power of advice-giving for self-motivation (Eskreis-Winkler et al.,2019,2018). The intervention was fully implemented through the official teacher job application platform before applicants viewed the list of schools and vacancies for application. Our intervention asked teachers in the treatment group applying to job vacancies to (voluntarily) advise a generic fellow applicant 2 –a hypothetical peer– about which types of schools they should choose if they wished to have the greatest social impact on students who needed them most. Teachers in the control group were asked to (voluntarily) answer two neutral questions about the application process. To rule out (or attenuate) potential priming effects (Weingarten et al.,2016), teachers in both the treatment and control groups initially received a brief explanation that outlined the different types of schools that exist in Peru (for example, schools where teachers could significantly impact student learning or receive career advancement benefits for working in rural or remote areas). In addition, icons appeared on the platform next to eligible schools, indicating which institutions offered teacher bonuses for working in disadvantaged areas (such as rural locations). Assuming that a considerable proportion of teachers have service-oriented or altruistic motivations –which seems reasonable, as shown by recent literature in the region (Ajzenman et al.,2024a,b)–, the advice given in this context could work through several channels. For instance, reducing cognitive dissonance (advising to apply to a hard-to-staff school while not doing so) or by prompting a concrete plan to maximize social impact (Rogers et al.,2015;Eskreis-Winkler et al.,2019). We document several results closely following our preregistered plan, including a preregistered heterogeneity analysis by gender, based on previous research by Ajzenman et al. (2024a) and Bertoni et al. (2023b) in the same context, showing that female candidates have a disproportionally higher preference for working in urban and more accessible schools. Focusing on our main specification, the proportion of male teacher candidates who ranked a hard-to-staff school as their top choice increased by 2.1 percentage points (pp) (significant at 5%; control mean: 57%) while there was no effect for female candidates (the difference between male and female candidates is significant at 5%). The proportion of male teacher candidates that ranked a hard-to-staff school as their second choice increased by 2.2 pp (significant at 5%; control mean: 57%) while there was no effect 3 for female candidates (the difference between male and female candidates is significant at 5%). These effect sizes align with previous findings in similar interventions. Ajzenman et al. (2024a) found that male teacher candidates’ likelihood of applying to disadvantaged schools increased between 3 and 3.4 pp, depending on the outcome. In a separate intervention, Ajzenman et al. (2024b) found that preferences for hard-to-staff schools increased by 1.4 to 5.2 pp, depending on the outcome. Although we cannot rule out other channels, we show evidence consistent with avoidance of cognitive dissonance. We document a strong correlation in the treatment group between the specific type of school teachers recommend and the type of school they end up applying to as their first choice. The proportion of hard-to-staff schools listed in the male teachers’ choice set was 0.9 pp higher (significant at 10%; control mean: 62%), while there was no effect for female candidates (the difference between male and female candidates was nonsignificant). The probability of being offered a position in a hard-to-staff school increased by 1.5 pp for male candidates (significant at 10%; control mean: 46%), while there was no effect for female candidates (the difference between male and female candidates was non significant). The differences across genders (and the fact that the nudge was mostly ineffective among female candidates) are not surprising and align with other papers showing the same pattern in the same context (Ajzenman et al.,2024a), plausibly explained by the fact that hard-tostaff schools are relatively farther away and female candidates tend to have less geographic mobility (see Ajzenman et al. (2024a) and Bertoni et al. (2022)). To further validate these results, we use causal forest techniques with honest splitting to examine heterogeneous treatment effects and identify teachers’ characteristics that maximize the effect of the intervention (Athey and Imbens,2016,2017)). Consistent with our manual estimates of heterogeneous effects, our analysis uncovers substantial heterogeneity in treatment effects but only in terms of gender. We designed the intervention based on recent insight in behavioral economics and psy4 chology: the giving advice effect. This effect suggests that individuals who give advice (rather than receive it) are more likely to make decisions that align with their own recommendations. In our experiment, we randomly assigned teachers to the treatment and control groups. Teachers in the treatment group were asked to recommend what types of schools their colleagues should prioritize if they wanted to maximize their impact on students’ learning and explain their recommendations. Teachers in the control group answered two neutral questions. Following Eskreis-Winkler et al. (2019), we hypothesized that prompting applicants to actively think about and advise colleagues on how to maximize their impact on students would increase the likelihood of choosing a hard-to-staff school. Our paper contributes to at least two strands of behavioral economics and education literature. First, it extends the research on teacher sorting and educational inequality. Extensive literature reveals that students from low-income backgrounds or with lower academic performance tend to be enrolled in schools with less qualified teachers (Boyd et al.,2006;Dieterle et al.,2015;Feng and Sass,2018;Lankford et al.,2002;Jackson,2009; Sass et al.,2012), which adversely affects their educational outcomes (Aaronson et al., 2007;Sass et al.,2012;Thiemann,2018). Research on strategies to mitigate teacher sorting has predominantly focused on monetary incentives, which, as evidenced by numerous studies (Clotfelter et al.,2008;Falch,2011;Glazerman et al.,2012;Springer et al.,2016; Rosa,2017;Bueno and Sass,2018;Feng and Sass,2018;Elacqua et al.,2019), often yield only modest or non significant impacts on teachers’ preferences for hard-to-staff schools (Neilson et al. (2021) being a noteworthy exception). In line with Ajzenman et al. (2024a) and Ajzenman et al. (2024b), we contribute to this literature by exploring a novel behavioral intervention to reduce teacher sorting at zero cost. Second, our paper intersects with an expanding literature on advice-giving. While most papers in this literature focus on the effects of receiving advice as a way of driving 5 The objective was twofold: to inform teachers about the meaning of the icons and to prime all participants (treated and control teachers) to think about hard-to-staff and remote schools. These icons were universally displayed to all teachers, irrespective of their treatment condition, as shown in Figure A2. In our sample, hard-to-staff schools were designated by two of these three icons: one for monetary bonuses and another for accelerated career progression due to disadvantaged conditions. This arises because hardto-staff schools in regional areas receive both benefits, while the vast majority of schools in Lima and Callao do not. However, to encourage teachers to apply to certain urban schools, the government needed a signaling mechanism, even if these schools did not receive any bonus. Since, in theory, teachers could potentially apply to positions in Lima and Callao under specific exceptions, we used the same initial icon screens across the country for consistency. After teachers passed the first page, the system displayed two voluntary questions. Teachers in the treatment group were asked to give advice to a hypothetical colleague who wants to maximize their impact on students’ learning. The first multiple choice question specifically asked what type of school they would recommend from a list of five: Vulnerable urban, VRAEM, Intercultural bilingual, Rural, or Non-vulnerable urban. In the open-ended question teachers were asked to justify their answer. The first four categories (Vulnerable urban, VRAEM, Intercultural bilingual, Rural) are considered hardto-staff and thus, in our sample, receive a bonus and are identified with the relevant icons. Importantly, those characteristics (not only the icons, but also the type of schools) were visible in the list of schools in both groups, so teachers could recognize them easily. Teachers in the control group received two neutral questions. In the first (multiple choice), they selected the outlet through which they received the most information about the process, from a list of five: Text messages, Email, Helpline, UGEL. In the second (open-ended), teachers wrote about the best outlet to receive information related to the 12 process. By design, teachers in both groups received exactly the same information and, within region, had access to the same list of vacancies and relevant information about schools. Section A1 of the Appendix provides a detailed description of the questions and illustrates how they were presented on the platform. The treatment draws inspiration from behavioral economics studies examining the impact of giving advice (primarily Eskreis-Winkler et al. (2018), Eskreis-Winkler et al. (2019)). Our hypothesis posits that by explicitly outlining strategies for teachers to make a social impact, particularly in teaching hard-to-staff students, those with prosocial tendencies—teachers who prioritize making a societal difference—would be inclined toward consistent behavior. Consequently, they would be more likely to apply to schools where their influence could be maximized. 4 Data This paper uses administrative data from the 2022 public school teacher selection process in Peru. The data include a comprehensive database of the vacancies offered in the teacher selection process. This database contains various school characteristics, such as location (region, province, district, UGEL), area (urban/rural), school type (multi-teacher, multigrade, or single-teacher), and indicators showing whether the school is bilingual, located in the VRAEM area or a frontier region, and whether it is classified as hard-to-staff. The dataset also contains information on the candidates evaluated in the centralized stage, specifically their performance on the standardized written exam. The dataset includes details about those assessed in the decentralized stage, including their results on both the pedagogical competence and professional trajectory assessment. Additionally, it contains information on candidates who successfully passed both stages, including their 13 ranked preferences within their selected region and the school where they were ultimately assigned. We restrict the sample to candidates applying for positions in the regular education track, known as Educaci´on Basica Regular (EBR). Throughout the analysis, we focus on two samples. The first is the ”full sample,” which includes all candidates who selected vacancies through the government platform in any region of Peru (except Lima Metropolitana and Callao, where there were virtually no hard-to-staff schools, as defined by the government). This sample consists of 74,692 individuals. Although the experiment was successfully implemented, many teachers were likely exposed to no variation in vacancy type, rendering the treatment ineffective. Teacher labor markets tend to be highly localized (Boyd et al.,2005;Reininger,2012;Bertoni et al.,2020). This is because teacher candidates likely select vacancies near their area of residence. Many candidates live in areas where there is no variation in the type of vacancy within their regions: either all vacancies were in hard-to-staff schools or non-hard-to-staff schools. We anticipated this problem, as it had occurred in the previous (Ajzenman et al., 2024a) and similar (Ajzenman et al.,2024b) contexts. Thus, we pre-registered the results of restricting the full sample to increase power, although we did not detail a specific procedure. This lack of variation in vacancy types arose because more than 20% of all districtsubject specialty area pairs had no hard-to-staff schools, and over half of district-specialty area pairs had only hard-to-staff schools. Teachers searching for vacancies within their district or nearby districts must select the available vacancy type based on their subject area, which made the treatment ineffective in cases where all vacancies were either hard-tostaff or non-hard-to-staff. For instance, 10.9% of teachers only had access to hard-to-staff schools in their residence district or nearby districts, based on their subject area. Similarly, 15.6% only had access to non-hard-to-staff schools. 14 Hence, following Ajzenman et al. (2024b), the second sample consists of a ’likely treated sample’. This sample includes teachers who have at least one hard-to-staff and one nonhard-to-staff school available in their district of residence (where they took the qualifying exam) or nearby districts. Given that the area of residence is pre-determined at the moment of entering the platform for the first time, this restriction is exogenous to the treatment, thus preserving the causal interpretation of the results. This likely treated sample consists of 45,578 individuals. Table 1 presents a descriptive summary of the applicants in both our restricted (i.e., sample with local supply restriction) and full samples. 15 Table 1: Summary of model variables Sample with local supply restriction Sample without local supply restriction Mean SD p10 p50 p90 Count Mean SD p10 p50 p90 Count Woman 0.77 0.42 0.00 1.00 1.00 45,578 0.71 0.46 0.00 1.00 1.00 74,692 Age 39.83 7.40 30.00 40.00 49.00 45,578 39.84 7.48 29.00 40.00 49.00 74,692 Disabled 0.01 0.12 0.00 0.00 0.00 45,578 0.02 0.12 0.00 0.00 0.00 74,692 Total score of centralized stage 140.87 17.51 119.00 139.00 166.00 45,578 140.72 17.59 119.00 139.00 166.00 74,692 Total score of decentralized stage 62.20 11.33 47.50 62.00 75.50 45,578 62.37 11.50 47.74 62.50 76.00 74,692 Selects top 1 0.58 0.49 0.00 1.00 1.00 45,578 0.63 0.48 0.00 1.00 1.00 74,692 Selects top 2 0.66 0.47 0.00 1.00 1.00 45,578 0.70 0.46 0.00 1.00 1.00 74,692 Selects top 3 0.72 0.45 0.00 1.00 1.00 45,578 0.75 0.44 0.00 1.00 1.00 74,692 Selects top 4 0.75 0.43 0.00 1.00 1.00 45,578 0.78 0.42 0.00 1.00 1.00 74,692 % of HTS until top 2 0.58 0.45 0.00 0.50 1.00 45,578 0.63 0.44 0.00 1.00 1.00 74,692 % of HTS until top 3 0.58 0.43 0.00 0.67 1.00 45,578 0.63 0.43 0.00 1.00 1.00 74,692 % of HTS until top 4 0.58 0.42 0.00 0.75 1.00 45,578 0.63 0.41 0.00 0.75 1.00 74,692 Assigned 0.71 0.46 0.00 1.00 1.00 45,578 0.68 0.47 0.00 1.00 1.00 74,692 Assigned HTS 0.46 0.50 0.00 0.00 1.00 45,578 0.48 0.50 0.00 0.00 1.00 74,692 Notes:Woman: Takes 1 if the teacher is a woman. Age: Age of the teacher. Disabled: Indicates whether the teacher has a disability.Total score of centralized stage: Score of the centralized stage(PUN exam).Total score of decentralized stage: Score of the decentralized stage. Selects top i: Takes 1 if at least one HTS is selected between ranking 1 and ranking i. % of HTS until top i: % of HTS selected until ranking i. Assigned: Takes 1 if the teacher was assigned to a vacancy. Assigned to HTS: Takes a 1 if the teacher was assigned to a vacancy on a HTS school. The description of the sample with local supply restriction is in Section 4 16 5 Empirical strategy To measure the overall impact of advice-giving on different teachers’ selection outcomes, we run regressions of the following form: Yi=α+βTi+γXi+ϵi(1) were Yiis the preference or assignment outcome for teacher candidate i.Tiis a dummy that indicates whether candidate ireceived the treatment, and Xiis a vector including candidate control variables, such as age, gender, region, grupo de inscripcion, and PUN score. Table 2 compares candidate characteristics in the between treatment and control groups for both samples. Most of the variables do not show any significant difference, and even those that do are quantitatively almost identical. 5.1 Main Measures Our analysis includes our (pre-registered) outcomes and pre-registered heterogeneity by gender. The outcomes relating to teacher choices are (i) if their first choice was a hardto-staff school; (ii) if their first two choices included a hard-to-staff school; and (iii) the proportion of hard-to-staff schools (as defined by the government) in teachers’ choice sets. Importantly, our main outcomes related to choice reflect high-stakes decisions; out of the teachers who are assigned to a school, 40.9% get their first preference and 51.7% get their top-2 preferences. We complement our analysis with an ”assignment” outcome that captures whether a candidate was offered a teaching job at a hard-to-staff school. The outcome ”Assigned to hard-to-staff school” takes the value of one if the candidate was assigned to work in a school categorized as ”hard-to-staff” by the algorithm. The assignment outcome should 17 Table 2: Balance test Sample with local restriction Sample without local restriction Control Treatment Treatment - Control Control Treatment Treatment - Control Woman 0.767 0.777 0.010** 0.702 0.708 0.006* (0.422) (0.416) [0.016σ] (0.457) (0.454) [0.010σ] Age 39.816 39.840 0.024 39.841 39.834 -0.008 (7.381) (7.409) [0.002σ] (7.477) (7.485) [-0.001σ] Disabled 0.013 0.015 0.001 0.015 0.016 0.000 (0.115) (0.120) [0.007σ] (0.123) (0.124) [0.001σ] General skills exam score 35.774 35.706 -0.068 35.931 35.814 -0.117** (7.417) (7.435) [-0.006σ] (7.416) (7.415) [-0.011σ] Specialization and pedagogical exam score 105.213 105.054 -0.159 104.946 104.754 -0.192* (14.572) (14.586) [-0.008σ] (14.537) (14.510) [-0.009σ] Total score of centralized exam 140.987 140.760 -0.227 140.877 140.568 -0.309** (17.488) (17.523) [-0.009σ] (17.591) (17.582) [-0.012σ] Specialty knowledge score 41.534 41.562 0.028 41.527 41.538 0.011 (6.186) (6.197) [0.003σ] (6.149) (6.142) [0.001σ] Personal interview score 8.217 8.244 0.027* 8.186 8.207 0.020* (1.568) (1.563) [0.012σ] (1.560) (1.552) [0.009σ] Total score of decentralized stage 62.182 62.216 0.034 62.373 62.359 -0.014 (11.259) (11.404) [0.002σ] (11.471) (11.537) [-0.001σ] Observations 22,796 22,782 45,578 37,296 37,396 74,692 Notes:Woman: Takes 1 if the teacher is a woman. Age: Age of the teacher. Disabled: Indicates whether the teacher has a disability. General skills exam score: Score from the first part of the centralized stage, related to general skills. Specialization and pedagogical exam score: Score from the second part of the centralized stage, focused on pedagogical, curricular, and disciplinary knowledge. Total score of centralized stage: Total score from the centralized stage (PUN exam). Specialty knowledge score: Score from the first part of the decentralized stage, which involves classroom observation. Personal interview score: Score from the second part of the decentralized stage, based on a personal interview. Total score of decentralized stage: Total score from the decentralized stage. The description of the sample with local supply restriction is in Section 4. *** p<0.01, ** p<0.05, * p<0.1. Standard errors are reported in parentheses below each mean, while standardized differences are reported below the difference columns. be interpreted with caution, as the effect of the experiment on teachers’ preferences in the treatment group could influence the outcome allocation of the control group in equilibrium (see Ajzenman et al. (2024a,b) for a discussion). Consistent with other papers in the same context (Ajzenman et al.,2024a) we use teachers’ PUN score (their grade on the qualifying exam) as a proxy for teacher quality or ability. While we acknowledge that teachers’ value-added is a better predictor of teachers’ performance (and, more generally, that teachers’ effectiveness depends on a variety of traits as shown by Rockoff et al. (2011); Jacob et al. (2018)), we were not able to use this indicator because most of the teacher candidates were new to the school system. 18 Reassuringly, Bertoni et al. (2023a) shows that, in Peru, PUN scores are significantly correlated with measures of value-added. 6 Results and Interpretation We display our main results in Table 3. For each outcome, we show two panels. Panel A - our preferred specification - shows the results using the likely treated sample (as defined in section 4) and Panel B uses the full sample. Columns labeled with odd numbers show the results for the main regression without interactions, and columns labeled with even numbers show the results including a gender dummy that takes the value of one for female candidates and zero otherwise. This heterogeneity was pre-registered and is based on previous research in the same context, documenting disproportionally higher preferences of female candidates for working in urban areas (see Ajzenman et al. (2024a) and Bertoni et al. (2022)). As columns (1) and (2) show in panel A, the intervention successfully increased the proportion of hard-to-staff schools included in teachers’ choice sets. For male and female candidates pooled, the effect is 0.4 percentage points (significant at 10%; mean in the control group 62.5%). Column (2) shows that the effect seems to be driven by male candidates: 0.7 pp (significant at 10%; mean in the control group 70.3%), although the gender interaction is not significant. Columns (4) to (6) confirm these results and emphasize the role of male teachers in driving the main effects. While there is no treatment effect on the probability of ranking a hard-to-staff school in the top-1 or top-2 positions for female and male candidates pooled, the effect for male candidates is significant and large. They are 2.1 pp. (significant at 5%; mean in the control group 64.6%) and 2.2 pp (significant at 5%; mean in the control group 65.3%) more likely to rank a hard-to-staff school as their top or 2nd best choices, respectively. Importantly, the 19 gender interaction becomes significant and negative in both cases, resulting in a null effect for female candidates. Considering that 51.7% were assigned to one of their top-2 choices, these outcomes are particularly relevant, as they refer to a high-stakes, consequential decision. Consistent with these results, columns (7) and (8) show that the probability of being assigned to a hard-to-staff school is 0.8 pp. (male and female candidates pooled) and 1.5 pp. for male candidates (although the gender interaction is not significant), respectively. The effects in both cases are significant at the 10% level (mean in the control group: 45.97% and 52.5% respectively). The primary objective of the intervention was to improve both the quantity and quality of teachers working in hard-to-staff schools. Since these schools are inherently difficult to staff and every teacher participating in this program passed a rigorous qualifying examination, the goal would have been accomplished even if the treatment impacted only the relatively lower-performing teachers in the sample. Ideally, an even more equitable outcome would have been if not only the lowest-performing teachers were influenced by the treatment. In Table 4 we analyze the heterogeneous treatment effects by candidate performance (using the qualifying test score as a continuous measure of future performance). To simplify the interpretation, we present separate results for female and male candidates. Our results suggest that the effect was not driven by low performers. Instead, the effect is larger for higher-performing teachers (Column 1).6While our quality measure is a proxy and may not be a strong predictor of teachers’ value-added, other papers in the same context show 6Even if the Deferred Acceptance (DA) mechanism were not strategy-proof, this table shows that the treatment does not reinforce strategic behavior in the expected way. Under a manipulable mechanism, the optimal response would likely involve high-scoring candidates avoiding HTS schools to secure positions in more desirable schools, knowing they would be prioritized for more competitive vacancies, while lower-scoring candidates would have stronger incentives to apply to HTS schools, anticipating lower competition. If the treatment simply made this strategic behavior more salient, we would expect the interaction between Treatment ×PUN Score to be negative, as lower-scoring candidates should respond more strongly. However, the positive (or null) coefficient contradicts this expectation. 20 a strong correlation between PUN and value-added (Bertoni et al.,2023a). Finally, given that the primary motivation for the intervention was to improve access to high-quality teachers for students enrolled in hard-to-staff institutions, we examined whether, as a result of the treatment, teachers selected and were offered positions in schools where students had relatively lower scores. Table 5 shows the results for the two students’ test scores: Math and Language. Each outcome reflects the average test score for each school in the corresponding subject. Specifically, the outcome related to top-1 is the average students’ test score in Math of the school chosen as the first priority in teachers’ choice set. Although the results are noisy, they show that teachers in the treatment group were offered positions in schools where, on average, students have significantly lower performance in Math and Language (and thus, students likely to be most in need). 21 ity. Intercultural bilingual schools received highest priority given their unique pedagogical demands and distinct instructional challenges, including specialized language requirements and cultural adaptations. VRAEM and frontier schools were assigned second-order priority, reflecting their operation in higher-risk zones characterized by socioeconomic vulnerability. Rural designation served as the residual category, encompassing schools that, while located in rural areas, did not present the specialized features of either bilingual or VRAEM/frontier institutions. This hierarchical classification ensures mutual exclusivity while preserving the most salient characteristics of each school type. Our classification approach rests on the assumption that teachers’ school selection reflects a hierarchical preference structure based on schools’ most distinctive attributes. When choosing intercultural bilingual schools, teachers likely prioritize the linguistic and cultural dimensions of instruction over geographic location. Similarly, the selection of VRAEM or frontier schools suggests that teachers weigh institutional vulnerability and challenging conditions more heavily than rural status per se. This assumption aligns with a decision-making framework where teachers evaluate schools based on their most distinguishing characteristics rather than their more general features. Such hierarchical preferences provide theoretical justification for our mutually exclusive categorization. 7 Discussion This paper underscores the critical problem of global teacher sorting in education systems and presents an experimental assessment to mitigate this issue in the Peruvian teacher selection process. By leveraging the behavioral principle of the giving advice effect, the study prompts teachers to offer recommendations about schools they should prioritize to impact students’ learning. The results contribute to the understanding of teacher sorting, educational inequality, and the efficacy of behavioral interventions in reducing 28 such disparities at no additional cost. Addressing teacher sorting through these interventions holds promise in promoting equity in education by attracting qualified teachers to underprivileged schools. This is particularly vital considering the substantial influence of effective educators on student outcomes, especially among those from disadvantaged backgrounds. Additionally, our findings indicate that scalable, cost-effective behavioral strategies can play a pivotal role in diminishing educational inequities by alleviating teacher shortages in less privileged schools. 8 Data Availability Statement The data supporting this study’s findings belong to the MINEDU and were used under specific license for this work; therefore, they are not publicly available. The data are, however, available from the authors upon reasonable request and with permission from MINEDU. 29 References Aaronson, Daniel, Lisa Barrow, and William Sander, “Teachers and student achievement in the Chicago public high schools,” Journal of Labor Economics, 2007, 25 (1), 95–135. Ajzenman, Nicol´as, Eleonora Bertoni, Gregory Elacqua, Luana Marotta, and Carolina M´endez Vargas, “Altruism or money? Reducing teacher sorting using behavioral strategies in Peru,” Journal of Labor Economics, 2024, 42 (4), 000–000. , Gregory Elacqua, Anal´ıa Jaimovich, and Graciela P´erez-N´u˜nez, “Humans versus Chatbots: Scaling-up behavioral interventions to reduce teacher shortages,” 2023. , , Luana Marotta, and Anne Olsen, “Order effects and employment decisions: Experimental evidence from a nationwide program,” 2024. Araujo, M Caridad, Pedro Carneiro, Yyann´u Cruz-Aguayo, and Norbert Schady, “Teacher quality and learning outcomes in kindergarten,” The Quarterly Journal of Economics, 2016, 131 (3), 1415–1453. Athey, Susan and Guido Imbens, “Recursive partitioning for heterogeneous causal effects,” Proceedings of the National Academy of Sciences, 2016, 113 (27), 7353–7360. and , “The state of applied econometrics: Causality and policy evaluation,” Journal of Economic Perspectives, 2017, 31 (2), 3–32. and Stefan Wager, “Estimating treatment effects with causal forests: An application,” Observational Studies, 2019, 5(2), 37–51. Bau, Natalie and Jishnu Das, “Teacher value added in a low-income country,” American Economic Journal: Economic Policy, 2020, 12 (1), 62–96. 30 Bertoni, Eleonora, Gregory Elacqua, Carolina M´endez, and Humberto Santos, “Teacher hiring instruments and teacher value added: Evidence from Peru,” Educational Evaluation and Policy Analysis (forthcoming), 2023. , , Diana Hincapi´e, Carolina M´endez, and Diana Paredes, “Teachers’ preferences for proximity and the implications for staffing schools: Evidence from Peru,” Education Finance and Policy, 2022, pp. 1–32. , , , , and Diana Paredese, “Teachers’ preferences for proximity and the implications for staffing schools: Evidence from peru,” Education Finance and Policy, 2023, 18 (2), 181–212. , , Luana Marotta, Matias Mart´ınez, Carolina M´endez, Veronica Montalva, Anne Sofie Westh Olsen, Sammara Soares, and Humberto Santos, “El problema de la escasez de docentes en Latinoam´erica y las pol´ıticas para enfrentarlo,” 2020. Bobba, Matteo, Tim Ederer, Gianmarco Leon-Ciliotta, Christopher Neilson, and Marco G Nieddu, “Teacher compensation and structural inequality: Evidence from centralized teacher school choice in Per´u,” Technical Report, National Bureau of Economic Research 2021. Boyd, Donald, Hamilton Lankford, Susanna Loeb, and James Wyckoff, “Explaining the short careers of high-achieving teachers in schools with low-performing students,” American Economic Review, 2005, 95 (2), 166–171. , , , and , “Analyzing the determinants of the matching of public school teachers to jobs: Disentangling the preferences of teachers and employers,” Journal of Labor Economics, 2013, 31 (1), 83–117. 31 , Pamela Grossman, Hamilton Lankford, Susanna Loeb, and James Wyckoff, “How changes in entry requirements alter the teacher workforce and affect student achievement,” Education Finance and Policy, 2006, 1(2), 176–216. Britto, Diogo GC, Paolo Pinotti, and Breno Sampaio, “The effect of job loss and unemployment insurance on crime in Brazil,” Econometrica, 2022, 90 (4), 1393–1423. Bueno, Carycruz and Tim R Sass, “The Effects of Differential Pay on Teacher Recruitment and Retention,” Andrew Young School of Policy Studies Research Paper Series, 2018, (18-07). Clotfelter, Charles, Elizabeth Glennie, Helen Ladd, and Jacob Vigdor, “Would higher salaries keep teachers in high-poverty schools? Evidence from a policy intervention in North Carolina,” Journal of Public Economics, 2008, 92 (5-6), 1352–1370. Dee, Thomas S and Dan Goldhaber, “Understanding and addressing teacher shortages in the United States,” The Hamilton Project, 2017. Dieterle, Steven, Cassandra M Guarino, Mark D Reckase, and Jeffrey M Wooldridge, “How do principals assign students to teachers? Finding evidence in administrative data and the implications for value added,” Journal of Policy Analysis and Management, 2015, 34 (1), 32–58. Elacqua, Gregory and Luana Marotta, “Is working one job better than many? Assessing the impact of multiple school jobs on teacher performance in Rio de Janeiro,” Economics of Education Review, 2020, 78. , Diana Hincapie, Isabel Hincapi´e, and Veronica Montalva, “Can Financial Incentives Help Disadvantaged Schools to Attract and Retain High-performing Teachers?: Evidence from Chile,” Technical Report, IDB Working Paper Series 1080, InterAmerican Development Bank November 2019. 32 Eskreis-Winkler, Lauren, Ayelet Fishbach, and Angela L Duckworth, “Dear Abby: Should I give advice or receive it?,” Psychological Science, 2018, 29 (11), 1797– 1806. , Katherine L Milkman, Dena M Gromet, and Angela L Duckworth, “A largescale field experiment shows giving advice improves academic outcomes for the advisor,” Proceedings of the national academy of sciences, 2019, 116 (30), 14808–14810. Evans, David K. and Amina Mendez Acosta, “How to recruit teachers for hard-tostaff schools: A systematic review of evidence from lowand middle-income countries,” Economics of Education Review, 2023, 95, 102430. Falch, Torberg, “Teacher mobility responses to wage changes: Evidence from a quasinatural experiment,” American Economic Review, 2011, 101 (3), 460–65. Feng, Li and Tim R Sass, “The impact of incentives to recruit and retain teachers in “hard-to-staff” subjects,” Journal of Policy Analysis and Management, 2018, 37 (1), 112–135. Glazerman, Steven, Ali Protik, Bing ru Teh, Julie Bruch, Neil Seftor et al., “Moving High-Performing Teachers Implementation of Transfer Incentives in Seven Districts,” Technical Report, Mathematica Policy Research 2012. Jackson, C. Kirabo, “Student demographics, teacher sorting, and teacher quality: Evidence from the end of school desegregation,” Journal of Labor Economics, 2009, 27 (2), 213–256. Jackson, C Kirabo, “What do test scores miss? The importance of teacher effects on non–test score outcomes,” Journal of Political Economy, 2018, 126 (5), 2072–2107. 33 Jacob, Brian A, Jonah E Rockoff, Eric S Taylor, Benjamin Lindy, and Rachel Rosen, “Teacher applicant hiring and teacher performance: Evidence from DC public schools,” Journal of Public Economics, 2018, 166, 81–97. Kane, Thomas J and Douglas O Staiger, “Estimating teacher impacts on student achievement: An experimental evaluation,” Technical Report, National Bureau of Economic Research 2008. Ladd, Helen F and Lucy C Sorensen, “Returns to teacher experience: Student achievement and motivation in middle school,” Education Finance and Policy, 2017, 12 (2), 241–279. Lankford, Hamilton, Susanna Loeb, and James Wyckoff, “Teacher sorting and the plight of urban schools: A descriptive analysis,” Educational Evaluation and Policy Analysis, 2002, 24 (1), 37–62. Marotta, Luana, “Teachers’ Contractual Ties and Student Achievement: The Effect of Temporary and Multiple-School Teachers in Brazil,” Comparative Education Review, 2019, 63 (3). Neilson, Christopher, Matteo Bobba, Tim Ederer, Gianmarco Leon-Ciliotta, and Marco Nieddu, “Teacher compensation and structural inequality: Evidence from centralized teacher school choice in peru,” 2021. Pop-Eleches, Cristian and Miguel Urquiola, “Going to a better school: Effects and behavioral responses,” American Economic Review, 2013, 103 (4), 1289–1324. Pugatch, Todd and Elizabeth Schroeder, “Incentives for teacher relocation: Evidence from the Gambian hardship allowance,” Economics of Education Review, 2014, 41, 120– 136. 34 Reininger, Michelle, “Hometown disadvantage? It depends on where you’re from: Teachers’ location preferences and the implications for staffing schools,” Educational Evaluation and Policy Analysis, 2012, 34 (2), 127–145. Rivkin, Steven G, Eric A Hanushek, and John F Kain, “Teachers, schools, and academic achievement,” Econometrica, 2005, 73 (2), 417–458. Rockoff, Jonah E, Brian A Jacob, Thomas J Kane, and Douglas O Staiger, “Can you recognize an effective teacher when you recruit one?,” Education finance and Policy, 2011, 6(1), 43–74. Rogers, Todd, Katherine L Milkman, Leslie K John, and Michael I Norton, “Beyond good intentions: Prompting people to make plans improves follow-through on important tasks,” Behavioral Science & Policy, 2015, 1(2), 33–41. Rosa, Leonardo, “Teacher Preferences in Developing Countries,” 2017. Sass, Tim R, Jane Hannaway, Zeyu Xu, David N Figlio, and Li Feng, “Value added of teachers in high-poverty schools and lower poverty schools,” Journal of urban Economics, 2012, 72 (2-3), 104–122. Schaerer, Michael, Leigh P Tost, Li Huang, Francesca Gino, and Rick Larrick, “Advice giving: A subtle pathway to power,” Personality and Social Psychology Bulletin, 2018, 44 (5), 746–761. Springer, Matthew G, Walker A Swain, and Luis A Rodriguez, “Effective teacher retention bonuses: Evidence from Tennessee,” Educational Evaluation and Policy Analysis, 2016, 38 (2), 199–221. Sutcher, Leib, Linda Darling-Hammond, and Desiree Carver-Thomas, “A coming crisis in teaching? Teacher supply, demand, and shortages in the US,” 2016. 35 Thiemann, Petra, “Inequality in education outcomes: The role of sorting among students, teachers, and schools,” Unpublished manuscript, Lund University. https://ekstern. filer. uib. no/svf/Econ% 20web/Petra% 20Thiemann teacher-sorting2018-10-16. pdf, 2018. UNESCO, “Global report on teachers: addressing teacher shortages,” International Task Force on Teachers for Education 2030, 2023, 1(1), 1–32. Weingarten, Evan, Qijia Chen, Maxwell McAdams, Jessica Yi, Justin Hepler, and Dolores Albarrac´ın, “From primed concepts to action: A meta-analysis of the behavioral effects of incidentally presented words,” Psychological bulletin, 2016, 142 (5), 472. 36 Appendix A1 Application Platform Following Ajzenman et al. (2024a), schools were labeled with icons highlighting their associated incentive schemes. Specifically, these consisted of a money bag icon, referencing the monetary incentives, a ladder, indicating the opportunity for faster career progression, and a school with a heart icon, symbolizing places where teachers could have a greater social impact. In both the treatment and control groups, right after teachers entered the platform, they saw a brief paragraph describing the next steps and pictures of the three icons that identified schools as hard-to-staff, as shown in Figure A1. These icons were shown to all teachers, regardless of the treatment condition, as shown in Figure A2. A1.1 Hard-to-staff schools in the platform A1.2 Intervention Before selecting vacancies on the platform, candidates answered two questions: one multiplechoice and one open-ended. The multiple-choice question for the treatment group was as follows: ”Many teachers committed our country’s education are participating in the 2022 CPM Entry Contest. If you had the opportunity to speak to one of them who expressed indecision about which educational institution to choose but knew they wanted to go to one where they could have the greatest impact on student learning, what type of educational institution would you recommend?” The options were: Vulnerable urban. Border or VRAEM, Intercultural bilingual, Rural, and Non-vulnerable urban. For the control group, the multiple-choice question was: ”Through which communication channel did you receive 37 A2 Causal forest Figure A7: Histogram of CATE: Selects HTS in first position (a) Likely treated sample (b) Full sample Notes: This figure shows the treatment effect heterogeneity obtained by the causal forest estimation in the outcome Selects HTS in first position.Selects HTS school in first position is a dummy that equals 1 if the applicants selected a hardto-staff school in their 1st position among their listed schools. The description of the sample with local supply restriction is in Section 4. 44 Figure A8: Histogram of CATE: Selects HTS in second position (a) Likely treated sample (b) Full sample Notes: This figure shows the treatment effect heterogeneity obtained by the causal forest estimation in the outcome Selects HTS in second position.Selects HTS school in second position is a dummy that equals 1 if the applicants selected a hard-to-staff school in their 2nd position among their listed schools. The description of the sample with local supply restriction is in Section 4. Figure A9: Histogram of CATE: % of HTS (a) Likely treated sample (b) Full sample Notes: This figure shows the treatment effect heterogeneity obtained by the causal forest estimation in the outcome % of listed HTS schools.% of listed HTS schools is the percentage of hard to staff schools in the choice set. The description of the sample with local supply restriction is in Section 4. 45 Figure A10: Histogram of CATE: Assigned to HTS (a) Likely treated sample (b) Full sample Notes: This figure shows the treatment effect heterogeneity obtained by the causal forest estimation in the outcome Assigned to HTS.Assigned to HTS is a dummy that equals 1 if the applicant is assigned to a HTS school. The description of the sample with local supply restriction is in Section 4. 46 Table A1: Covariates mean for top and bottom quintiles of treatment effect Full Sample Restricted Sample % of listed HTS schools Q1 (I) Q5 (II) Difference (II)-(I) Q1 (I) Q5 (II) Difference (II)-(I) Man 0.262 0.272 0.010 0.196 0.225 0.029 [0.034] [0.133] Age 39.763 40.209 0.446 39.120 39.953 0.833 [0.011] [0.021] Disabled 0.015 0.015 0.000 0.011 0.012 0.000 [0.011] [0.031] Specialty knowledge 101.505 109.386 7.881 101.306 111.263 9.958 [0.075] [0.095] Centralized exam score 137.312 146.828 9.515 136.242 148.959 12.717 [0.068] [0.090] Personal interview score 8.428 8.200 -0.228 8.152 8.294 0.142 [-0.028] [0.017] Selects HTS school in 1st position Man 0.224 0.298 0.074 0.166 0.243 0.077 [0.257] [0.351] Age 38.744 40.891 2.147 38.547 40.603 2.057 [0.054] [0.052] Disabled 0.014 0.016 0.002 0.011 0.015 0.003 [0.126] [0.241] Specialty knowledge 103.455 106.641 3.186 102.495 107.254 4.759 [0.030] [0.045] Centralized exam score 140.295 143.075 2.780 138.779 143.199 4.420 [0.020] [0.031] Personal interview score 8.337 8.297 -0.040 8.067 8.416 0.348 [-0.005] [0.042] Selects HTS school in 2nd position Man 0.240 0.280 0.040 0.166 0.220 0.054 [0.139] [0.244] Age 39.520 40.289 0.769 39.435 39.464 0.029 [0.019] [0.001] Disabled 0.014 0.016 0.001 0.012 0.014 0.002 [0.082] [0.131] Specialty knowledge 102.353 108.606 6.253 102.311 109.332 7.021 [0.060] [0.067] Centralized exam score 138.811 144.991 6.180 138.282 144.895 6.614 [0.044] [0.047] Personal interview score 8.420 8.190 -0.229 8.325 8.333 0.008 [-0.028] [0.001] Teacher is assigned to a HTS school Man 0.254 0.293 0.039 0.181 0.216 0.035 [0.135] [0.160] Age 40.006 38.392 -1.614 39.650 38.869 -0.781 [-0.040] [-0.020] Disabled 0.015 0.014 -0.001 0.013 0.014 0.000 [-0.093] [0.031] Specialty knowledge score 103.140 107.437 4.297 104.954 106.293 1.339 [0.041] [0.013] Centralized exam score 138.886 144.657 5.771 140.816 142.481 1.665 [0.041] [0.012] Personal interview score 8.306 8.359 0.053 8.346 8.326 -0.020 [0.006] [-0.002] Notes: This table shows the mean value of each covariate for the bottom and top quintiles of the treatment effect. The difference columns display the differences in points and also as differences expressed in standard deviations of the variable. The outcome % of listed HTS schools is the percentage of hard to staff schools in the choice set. The outcome Selects HTS school in is a dummy that equals 1 if the applicants selected a hard-to-staff school in their 1st or 2nd position among their listed schools, respectively. Teacher is assigned to a HTS school is a dummy that equals 1 if the applicant is assigned to a HTS school. The description of the sample with local supply restriction is in Section 4. 47