Village Fairness Norms and Land-Rental Markets

Abstract

This paper documents the role of village fairness norms in land markets. A strong and robust relationship is established between experimentally elicited village-level fairness norms and land-rental rates across 250 Malawian villages. Stronger fairness norms correlate with a tighter range in village rental rates. The study suggests that the fairness norms for tenants appear to be more important, constraining the land-rental price range by a price ceiling rather than a price floor. The results further indicate that rented-in fields are of lower agronomic quality than owner-cultivated fields, but do not find any statistically significant relationship between the fairness norms and land-rental activity in the village.

1. Introduction

Across Sub-Saharan Africa, farmers cultivate small fields in regions where land tenure systems are customary and communal, meaning that the land is owned or managed by kin-based or local governing bodies. Individuals obtain rights to cultivate such land through membership in those groups (Udry 2011). These customary and communal land tenure systems can result in flexible and negotiable arrangements for land tenure (Boserup 1985; Takane 2008) in which social norms or local values mediate land rental and purchase prices in ways competitive market systems do not always recognize or countenance.

This paper explores how social norms, specifically norms that compel communities towards “fair,” or equal, allocations, correlate with important properties of land-rental markets. In a simple supply-demand model of a land-rental market governed by such social norms, the norms might act as a minimum or maximum land-rental price, and hence lead to excess supply or excess demand.¹ This paper is set in rural Malawi where, as Kishindo (2004) notes, local chiefs in charge of allocating customary land have been charged with ensuring “equitable distribution of the land among the current generation and its preservation for future generations” (Saidi 1999).² Leader-driven allocation of land is subject to heterogeneous approaches reflecting, potentially, differential norms across villages.

Results from focus group interviews conducted in eight villages prior to the start of the larger study suggest both the importance of social factors and the presence of important variation in norms across villages.³ For example, when asked about land-rental rates, respondents noted that a considerable amount of the variation across villages largely related to location and population pressures. They further explained that there was little variation in land-rental rates within villages. Even after prompting, respondents in all but two villages remained adamant that the rental rates do not correlate with soil fertility or other favorable attributes of the field. Instead, they described the importance of social factors, reporting that landlords charged lower rents for relatives, for people in need, and for tenants using the land for subsistence food crops. Often, respondents had difficulty imagining what would happen if a landlord set a rental rate higher than the “correct” price for their village. At the same time, they noted the presence of excess demand: farmers would like to rent land in at this price in their village but cannot find a field. This lack of within-village variation limits incentives for owners of agronomically good fields, and may constrain rental markets.

This paper uses original survey and experimental data from 250 villages in rural Malawi to explore the relationships between norms and rental markets for land and to work through potential economic implications. The data link a measure of village fairness norms with land-market participation data from 2,500 farming households and 4,652 associated agricultural fields.

The analysis is descriptive and the documented patterns inform a framework that invites alternative interpretations of observed phenomena in land markets across Sub-Saharan Africa. Results in this paper are neither definitive nor causal. Instead, results suggest that future research might work to develop theoretical and empirical models that can more rigorously examine how fairness norms influence land markets in similar settings.

The paper’s analysis has three components. The first part of the analysis documents the correlates of land-rental rates in the data. A positive correlation between land rents and measures of marginal productivity is expected under a competitive market equilibrium. Instead, results show that field attributes and village characteristics explain merely 30 percent of the observed variation in land-rental rates.⁴ Furthermore, the results show that fields with markedly different soil and plot characteristics often feature the same rental price within a given village, and distributions of important characteristics are often at odds with distributions of rental rates. For example, while soil carbon, a key measure of soil quality, is approximately normally distributed, rental rates are not. Instead, the distribution of village rental rates appear to be capped with focal points. Some villages exhibit bunching at the lower end of the distribution of rental rates, suggesting the presence of a rental-rate floor, protecting landlords. A majority, however, exhibit upper-end bunching, indicating the possible presence of a maximum rental rate, protecting tenants.

Second, we use an innovative game to elicit village fairness norms. We form two groups of residents in each village, each tasked with choosing between two payoff distributions via majority vote. One distribution provides equal income to all members of either group, the second provides unequal, but Pareto-improving, payoff distributions, where each member of one group receives a larger payoff than members of the other group. The results show that groups on average demonstrate a preference for the equal and inferior aggregate payoffs. Of the 500 groups, 67.4 percent chose the inferior distribution when they would receive the smaller payoff and 34.4 percent chose the inferior distribution when they would receive the larger payoff. We link responses in this game to subjective equity preferences for equal land distribution elicited through a household survey. In this survey, 78 percent of respondents reported it is (very) unfair to increase the rent for one’s regular tenant, and 90 percent said it is (very) unfair to increase the rent of a tenant in need even though land is scarce. The results show that there is a strong correlation between the strength of the experimentally elicited village fairness norm and the share of respondents that answered (very) unfair to 8 out of 12 questions of the survey-elicited preferences. This correlation is embedded in a significant degree of between-village variation in both fairness norms and equity preferences.

The last step in the analysis demonstrates that this relationship between preferences and equity preferences extends beyond the setup of the experimental game and into the analysis of village land-rental rates and rental-market activity. Results of regressions of village-level fairness norms against the range of village-level rental rates show that stronger fairness norms correlate with a reduction in this rental-rate range. These results are robust to the inclusion of village-level controls and to multiple approaches to measuring the variability in the rental rate. Disaggregating the analysis by separately assessing the relationship between the strength of the norm, the minimum rental rate, and the maximum rental rate, provides further insight into the role of norms. The results indicate a negative correlation between the maximum rental rate and the strength of a village’s fairness norm, suggesting that the norm may serve primarily to protect tenants from rent increases. These empirical results are consistent with insights from the focus group interviews revealing prevailing concern for tenants, as well as the descriptive analysis of survey questions related to equity preferences.

As the data are limited in geographic and temporal scope (representing the situation in 2018 in Central Malawi) this study cannot analyze the dynamic evolution of fairness norms nor can results speak to their origins. This study is also unable to characterize how such norms are correlated with other unobservable features within and across villages that could generate spurious correlations. The approach of this paper hence is to demonstrate the robustness of the correlations observed and to suggest that fairness norms may constrain markets in this context, keeping prices away from their competitive equilibrium levels.

This paper documents a pattern of behavior that fits the conventional definition of a social norm, that is, an informal standard of acceptable practice within a group or community. However, one cannot make a claim as to whether one is analyzing the influence of commonly held preferences or group-imposed social norms (Burke and Young 2011; Bowles 2006). These two could be mutually reinforcing without a clear causal pathway from one to the other.⁵

We assume that the norm we measure is continuous in nature, ranging in terms of the share of individuals in the community who prefer equity outcomes to efficiency. Even so, the influence of norms is observed here in a cross sectional data and therefore we remain somewhat agnostic regarding whether the norm observed and measured is weakening, strengthening, or at its equilibrium state. Our approach is consistent with Axelrod (1986) and Binmore (1998), who model norms as evolving in steps and stages over time, the outcome of many individual decisions in the context of a repeated game. The norm is what moves the village away from the “economically efficient” outcome in terms of land-rental rates being allowed to rise to their market-driven level.⁶

Irrespective of the origins and exact nature of these norms, a conception of land distribution that internalizes the presence of fairness norms helps to explain stylized facts in Malawi, common across Sub-Saharan Africa. For example, land-rental markets are relatively thin in Malawi (Tione and Holden 2020), and the potential benefits of a thicker land-rental market are estimated to be high. According to Malawi’s Integrated Household Panel Survey (IHPS 2010, 2013, 2016) approximately 17 percent of rural households participate in land-rental markets annually.⁷Restuccia and Santaeulalia-Llopis (2017) estimate that Malawi’s aggregate agricultural output could increase by a factor of 3.6 if landholdings were reallocated to their most efficient use among existing farmers.⁸ Fairness norms can explain why these types of land allocations would not emerge via a market mechanism. It may be that land markets remain thin because norms constrain landlords from setting prices that would be more appealing than alternative uses for their land.⁹

The paper contributes to the literature on land markets in low-income countries, a long focus of development economists and policymakers. A substantial literature examines the role of land tenure security in agricultural development, household production and technology adoption, and welfare and migration (Goldstein and Udry 2008; Bellemare, et al. 2003; Deininger and Feder 2001; Gebremedhin and Swinton 2003; Ali, Dercon, and Gautam 2011; Deininger and Jin 2006; Mullan, Grosjean, and Kontoleon 2011; Salmerón-Manzano and Manzano-Agugliaro 2023; Bambio and Agha 2018; Besley 1995; Giles and Mu 2018; Deininger, Savastano, and Xia 2017). Researchers have also studied the functioning of land-rental markets, characterizing a range of frictions including asymmetric information about land quality, high transactions costs, and incentive problems (Binswanger and Rosenzweig 1986; Jin and Jayne 2013; Ricker-Gilbert and Chamberlin 2018; Binswanger, Deininger, and Feder 1993; Chamberlin and Ricker-Gilbert 2016; Abay, Chamberlin, and Berhane 2021). While research has focused on the way that land tenure can impact social outcomes (including conflict; see Peters (2004) and Deininger and Castagnini (2006) among others), few studies have examined the role of social processes and norms. This work suggests the potential importance of these processes.

The study also contributes to a body of research that explores how norms influence the functioning of a range of economic institutions (Michalopoulos and Xue 2021; Nunn 2020; Munshi 2014; Fafchamps 2011). This paper extends this type of analysis to land allocations. A primary focus of this work has been on labor markets in low-income countries (Goerges and Nosenzo 2020; Kaur 2019) but other evidence suggests that social norms affect, for example, the pricing of private property rights, conservation payments, and resource allocation (Chen et al. 2009; Kim 2007; Duflo and Udry 2004). While social structures and relations can affect agricultural productivity through insurance, credit, and information channels (Jakiela and Ozier 2016; Platteau 2000), limited studies have focused on the potential role of social norms in rural asset markets. Our paper’s results on the relevance of fairness norms to land-rental markets begins to fill that gap and suggest important new areas of research.

The paper proceeds as follows: Section 2 provides some background information on Malawi. Section 3 provides an overview of the data collected. Section 4 presents the descriptive analysis. Section 5 presents the regression analysis of land-market outcomes on fairness norms. Section 6 concludes.

2. Rural Malawi

Malawi is a landlocked country in Southern Africa with over 50 percent of the population living below the poverty line (FAO 2015). More than 80 percent of the country’s rural population derive their livelihoods directly from agriculture (Chinsinga 2011; FAO 2015). The country maintains three types of land tenure systems: private (freehold or leasehold), public, and customary (or traditional) (Ricker-Gilbert et al. 2019). Most farmland in rural Malawi is held under the traditional tenure system. Lunduka, Holden, and Øygard (2009) report that in the 1970s, about 80 percent of all arable land was held under the traditional tenure system, and by 1997, less than 10 percent of all arable land held under the customary system had been converted to private land. In this customary system, a household gains access to farmland either through direct allocation by the village chiefs, or through the inheritance of perpetual user rights. Kishindo (2004) notes that while Malawi’s pre-1994 Banda government treated customary land as a “reservoir from which private and public land can be obtained,” access to land was considered a right of all Malawians. The essential role of local chiefs (who were in charge of customary land) was to ensure “equitable distribution of the land among the current generation and its preservation for future generations” (Presidential Commission of Inquiry on Land Policy Reform 1999).

Farm plots in rural Malawi are small, with an average land-holding size of 0.75 ha in Central Malawi, and demand for land is increasing due to population pressure (Chamberlin and Ricker-Gilbert 2016; Deininger and Byerlee 2011). While land sales can provide a means of facilitating land transfer, the lack of tenure security renders this practice unusual in rural Malawi (Peters and Kambewa 2007; Peters 2013). Land-rental markets have emerged as a means of transferring land between households (Tione and Holden 2021; Holden and Otsuka 2014; Holden, Otsuka, and Place 2010). However, as noted, participation in these rental markets remains limited.

As a means of redistributing land from large estates to smallholders, as well as improving tenure security, Malawi adopted the National Land Policy in 2002 (FAO 2015). This policy sought to protect smallholders’ land rights and safeguard poor land users’ interests with the goal of facilitating the development of land-rental markets.¹⁰ The policy has been amended since its initial adoption. The most recent Customary Land Act, which took effect in 2018, fosters registration of customary land as private (FAO 2015). Early evidence indicates that this formalization process has increased land sales and rentals (Chamberlin and Ricker-Gilbert 2016).

Given Malawi’s recent policy interest in land markets and given persistently low rental-market participation, the country offers a relevant setting to study the role of social norms in the functioning and development of rental land markets.

3. Data Collected

The data for this study are from two districts in Central Malawi: Kasungu and Dowa. Within these districts, we selected two Extension Planning Areas (EPAs, a subdistrict administrative unit): Chibvala in the Dowa district and Mtunthama in the Kasungu district. We randomly sampled 250 villages¹¹ and 10 farming households living in each village, yielding 2,500 households in the sample. This sample was defined as part of a multi-year randomized controlled trial (RCT) focused on agricultural practices.¹² The measure of social norms employed is not based on this survey sample of 10 households per village, but a sample of 22 households per village (details on the latter are provided in the discussion of the dictator game).

We collected two rounds of panel data: one round in 2014 and a second in 2018. The analysis uses the 2018 round, as we collected norm measures only in that year.¹³ We collected quantitative, qualitative, and agronomic data from the village head, households, and fields, and conducted the game within each village. We administered a household survey with the household head, eliciting information on household assets, demographics, landholdings and land relations, distance from the homestead to household farmland fields, and equity preferences. For a randomly selected subset of the households, we also collected and analyzed soil data from the household’s primary field.

We administered a village-level questionnaire in each of the 250 villages with an individual knowledgeable about village functioning, often the village chief, in 2014 and 2018. The 2014 village survey collected information on factors known to play a role in land markets, such as the distance from the village to the closest input/output markets, the distance from the village to paved roads, and the number of households, among others. While the analyses rely mostly on data collected in 2018, this study uses village data from the 2014 round as control variables.

Table 1 summarizes the village-level characteristics. Panel 1 presents a summary of the game results (this is discussed in-depth later when the game is introduced). Panel 2 summarizes market access variables collected in the 2014 village survey, including distance from the village to a paved or all-weather road (km) and distance to the closest inputs market (km), as well as rates of in and out migration collected in the 2018 village survey, presence of irrigation in the village, and the number of households in the village. Panel 3 documents other village characteristics compiled from the household survey: ethnic composition, the share of farmers cultivating soy, an intra-cluster correlation coefficient for the per acre rental rates, and the share of villages with at least some rental activity in the 10 years preceding the 2018 survey.

3.1. Land Rents

We collected information about all fields owned and/or cultivated by the household. We asked whether the household participated in the land-rental market either as a tenant, landowner, or both at any point in the 10 years preceding the 2018 survey. This information is used to categorize each household by its longer-term land-rental market participation status, defining tenant households as those that exclusively rented in land at any point in the 10 years preceding the survey and landlord households as those that only rented out land at any point in the last 10 years preceding the survey. For all owned fields, we elicited the estimated sale value of the field (in Malawian kwacha, MK) and, importantly, the estimated yearly rental rate, also in MK.¹⁴ While traditionally this type of question is done using a willingness to pay and willingness to accept elicitation strategy, our framing appears to result in a measure which approximates the market prices (our formulation has also been used in the IHPS) (Julien, Bravo-Ureta, and Rada 2019; Ragasa and Mazunda 2018). This hypothetical phrasing was well understood by respondents in pre-testing; all respondents were familiar with the concept of land rentals and all the villages exhibit some degree of land-rental activity (see table 1).

For rented-in fields, we recorded the actual rental rate paid by the farmer. We similarly collected the actual rental rate for rented-out fields. The respondent had the opportunity to indicate whether this rental rate was paid as a share of the output or as a fixed amount of cash paid before planting or after harvest. We also asked the respondent to indicate the reference period for the rental agreement (yearly versus by season).¹⁵

Finally, we elicited responses regarding perceived soil texture, soil fertility, soil depth, field location, and cultivable and cultivated acreage. We randomly selected a subset of 521 households for whom we collected GPS-measured acreage and analyzed soil samples.¹⁶ We used standard sampling practices and collected cropping and management history for each field. Supplementary online appendix S3 describes soil sampling and lab testing methods.¹⁷

3.2. Inequity Aversion

We elicited inequity preferences by presenting hypothetical scenarios to the survey respondent which he or she was asked to rate as “very unfair,” “unfair,” “fair,” or “very fair” (allowing for a “no opinion” option). The scenarios covered relationships, responses to shocks, and changes in market demand and supply of land. For example, the scenario “Last year the rental rate was 15,000 kwacha. This year there are more tenants seeking to rent land. One landlord decides to charge an increased rate, 17,000 kwacha,” inquires about protection for tenants in the form of maximum rental rates. The scenario “A landowner usually rents out an acre of land for 15,000 kwacha. His son becomes sick, and the medical bills are very expensive. He increases the rent to 17,000 kwacha,” asks about protection for landlords. And the scenario “Last year the land rental rate was 15,000 kwacha per acre. This year, there is a new buyer in the market for soy and tobacco, who is offering prices up to 15 percent higher compared to last year. Landlords increase this year’s rent also by about 15 percent, from 15,000 kwacha to 17,000 kwacha,” inquires as to how the rental rate responds to market shocks.¹⁸

The full list of scenarios is included in supplementary online appendix table S1.1. These questions were inspired by Kaur (2019) who used a similar line of questions to obtain suggestive evidence on the relevance of fairness considerations in the casual labor market in India. Kaur (2019) was concerned with the circumstance of the relatively weaker party in the market—the laborer. By building questions on Kaur’s general design, this study focused on what is often thought of as the weakest party in the land-rental market—the tenant. While some of the scenarios were designed to capture preferences as they relate to the protection of tenants and others relate to protecting landlords, the questions were not explicitly designed to compare landlord-related norms versus tenant-related norms.

We did not explicitly inform respondents about whom these scenarios were unfair to. This could have influenced how respondents interpreted the person being treated unfairly as either the tenant or landlord depending on their perspective and participation status in the rental market. While it is possible that landlords in particular would act differently in actual, i.e. not hypothetical, scenarios, previous literature has indicated that answers to survey questions regarding social preferences tend to be quite stable, and incentivizing social-preference-related questions in the laboratory does not alter responses dramatically (see among others Falk et al. 2023; Chuang and Schechter 2015).

3.3. Dictator Game

We invited 22 individuals from 22 unique households to participate in the dictator game in each village. This number was decided on purposefully, so as to provide a reasonable representation of the village (the average village has 75 households; see table 1), and to avoid any ties in the subsequent votes during the dictator game.¹⁹ We recognize that sampling variation may affect the measure of social norms. However, as the sample/population ratio is high, and there is a good amount of within-village agreement, we expect this issue to be of limited importance in this case.

In a standard dictator game, a dictator divides a fixed budget between themselves and another, typically anonymous, participant. The amount shared with the other participant is commonly used as a measure of other-regarding preferences, whether a form of altruism or inequity aversion. As this study is interested in measuring the behavior of the group, we, therefore, modified this standard setup so that the participants played as a group, with the choice in the dictator game framed as a group choice. In each village, we divided all participants into two random groups and asked each group to make a series of choices (through a confidential majority vote).

In each round, a choice had to be made between two payoff bundles: Bundle A (500 MK; 500 MK) and Bundle B (500 MK; 1,000 MK). In the first bundle, both groups receive 500 MK per individual; in the second bundle, one group receives 500 MK per individual while another group receives 1,000 MK per individual. Note that Bundle B dominates Bundle A, and if post-game redistribution happens, this dominance could be strict. The group’s choice was settled through a majority vote, i.e., it was the choice of the majority of the group members. To frame the choice further as a group’s choice, we encouraged group members to communicate with each other prior to making their individual choices.²⁰

Each participant was given two colored cards (a blue card and a red card). The blue card represented a vote for Bundle A (500 MK; 500 MK) and red represented a vote for Bundle B (500 MK; 1,000 MK). Note that within the Malawian context, these colors are devoid of any particular significance. To obtain a group’s vote, we used a non-transparent cloth bag, in which participants could then drop the card of their preferred choice, either Bundle A or Bundle B. The non-transparency of the cloth bag, as well as the fact that we physically separated the participants when making their choice, ensured confidentiality. The researchers counted the votes of each group after each round. Note that the design guaranteed anonymity: even the researchers who recorded the total number of votes for each bundle did not know who in the group had voted for which bundle. This was important to avoid well-documented social pressures (see, e.g., Ambler, Godlonton, and Recalde 2021). The full game protocol is in supplementary online appendix S2.

Table 2 gives a summary of how the game was played in all three rounds. The three rounds differ based on which group receives the higher payoff in the case of Bundle B being selected. In the first round, if a group opted for Bundle B, then members in that group received 1,000 MK each, while members in the other group received only 500 MK per person. In the second round, if a group decided on Bundle B, then each member of the group received 500 MK, while members in the other group received 1,000 MK each. In the third round, if a group decided on Bundle B, the group to receive the higher amount (i.e. 1,000 MK each) was determined by a coin toss at the end of the dictator game (so the result of this coin toss was unknown at the time of the decision).

The dictator game captures what can be thought of as “reduced-form preferences,” meaning that they would account for any post-experiment redistribution. While redistribution, or risk-sharing, is common in communities in the developing world (see among others Dercon (2002) and Townsend (1995)), one can expect, in this case, redistribution post-game to be limited for two main reasons: (a) the amounts transferred by the dictator game are small, and (b) they do not represent a sustained stream of good or bad luck (Jakiela 2013). At the end of the dictator game, we implemented one of the group’s decisions from one of these three rounds, randomly chosen (out of the six possible options).²¹

3.4. Qualitative Data Collection

This study was motivated by a series of qualitative interviews done in the area. In 2016, we visited four villages in the Dowa/Kasungu districts and conducted focus group interviews with farmer clubs, following best practices (see Morgan 1996; Krueger and Casey 2015), and inquired about topics including perceptions of soil fertility, land sales, and land-rental markets. At the beginning of 2018, prior to this survey, we visited an additional 4 villages in the same region and conducted focus group interviews with about 10 randomly selected individuals. We also did four semi-structured qualitative interviews with key informants in the village (the village chief, if present). In these interviews, we further probed into perceptions of variation in soil fertility, drivers and participation in land-rental markets, and social norms. Prior to the start of the household survey, we also visited another 6 villages in the region to pre-test the data instruments. We conducted the interviews with 2 individuals (one male and one female), and pre-tested the game in groups of 22 randomly selected individuals in each village.

4. Descriptive Analysis

Table 3 reports household, field, and land-rental contract summary statistics for the full sample. Panel 1 reports sample household characteristics (namely sex, age, years of education of household head, household size, household access to credit, government subsidy, and value of household asset). Panel 2 reports on household land ownership and rental-market participation. Fifteen percent of households participated in the land market as tenants (in the 2017/18 agricultural season), while 5 percent participated as landlords (6 percent participated as both).

Using households’ responses about their participation in land markets over the previous 10 years, the sample is split into four groups. The table classifies each household as either (a) tenant, (b) landlord, (c) both tenant and landlord, or (d) neither.²² Note the discrepancy with the earlier numbers. While 34.5 percent of sampled households indicated they had rented in land in the 10 years previous to the survey, only 15 percent reported that they rented in land in the 2017/2018 agricultural season. Similarly, while 18 percent of the households reported that they rented out land at some point in the previous 10 years, only 5 percent reported renting out land in the 2017/2018 season.

Supplementary online appendix table S1.2 reports the means and standard deviations of household characteristics based on these sub-group categorizations. Again one notes the fluidity of the tenant/landlord definition. Households classified as tenants do not rent in every year, only 27 percent rented in in 2017/18. Among the households classified as tenants, only 20 percent rented out in 2017/18.²³ The results do not indicate sharp differences in group means across the majority of household characteristics (e.g., sex, age, and years of education of household head, household size), but find substantial differences in household access to credit across the groups. While 42 percent of tenant households indicated they have easy access to credit and 56 percent of households classified as tenant and landlord noted they have easy access to credit, only 37 percent and 33 percent of households classified as landlord and neither, respectively, indicated having easy access to credit. In addition, households who participate in the market as both tenants and landlords report relatively higher field sizes (2.40 acres).

Panel 3 reports on the summary of field characteristics. On average, households walk about 24 minutes to reach their fields. Farmers perceive approximately half of the fields as good quality. The data contain lab-based measures of soil quality for a subset of 571 of the fields; the average active carbon level is relatively low: 350 mg/kg. Agronomists generally consider soils with active carbon less than 350 to be poor and soils with active carbon greater than or equal to 700 to be excellent. Figure 1 presents the histogram of the distribution of active carbon in sampled soils across the villages. The skewness parameter for the distribution of this soil quality measure is 0.25 and the intra-cluster coefficient in land quality is 0.56. While significant, it also indicates that there is still substantial variation in soil quality within each village. This is consistent with other recent studies (Berazneva et al. 2018; Tittonell et al. 2005). As soil quality affects the agronomic potential of the field, this within-village dispersion in soil quality can potentially play an important role in pushing the dispersion of land-rental rates within the village (Wineman and Jayne 2018; Ricker-Gilbert and Mason 2017).

Panel 4 presents the rental rates and other contractual details. The majority of land-rental contracts are single-year, fixed-rent contracts, with payments made at the start of the primary rainy season, and 62 percent of rental contracts are within village.²⁴ Among households that rented in land, the average payment made was 20,255 MK per acre per year (USD 27). Table 3 notes an actual rental rate of about 20,000 MK per acre, compared to a hypothetical rate of about 23,000 MK per acre; a relatively small but statistically significant difference. In the analysis, we use the hypothetical rental rate which allows us to avoid small sample issues associated with using the actual rental rate (as few households rent land in or out in any given year).²⁵

Supplementary online appendix fig. S1.1 presents a histogram of the (hypothetical) rental rates. The rental rate ranges from 1,000 MK per acre to 80,000 MK per acre, per year; the distribution of rents is long-tailed (supplementary online appendix fig. S1.1). The analysis uses the ln-transformed rental rate presented in fig. 2. Within-village rental rates are often clustered at the lower end of the rental-rate histogram. See supplementary online appendix fig. S1.3 for the rental-rate distributions of nine randomly selected villages.²⁶

4.1. Rental-Rate Correlates

In equilibrium, absent any fairness norms, the village-level supply of land and village-level demand for land will determine the rental rate. Unlike sale prices, rental rates capture only the agricultural value of land, i.e., the annual returns of land in terms of marketed and consumed agricultural production. This, in turn, depends on the market conditions and the field characteristics: soil quality, plot area, location, and village-level characteristics including market access and climate conditions.²⁷

The results indicate that rental rates correlate to some extent with field attributes, capture soil quality and location, and village characteristics related to market access. Table 4 presents the results of a regression of the logarithm of the hypothetical per acre rental rate on a constant (columns 1 and 2), adding village location (columns 3 and 4), adding field characteristics (columns 5 and 6), and adding both village location and field characteristics (columns 7 and 8). Columns 1, 3, 5, and 7 present results without village-fixed effects; columns 2, 4, 6, and 8 present results with village-fixed effects.

The results presented in columns 3 and 5 suggest that smaller, good-quality fields with limited erosion located close to the homestead fetch a higher rent.²⁸ These field characteristics explain 14.7 percent of the variation in land-rental rates. Fields in villages further away from markets and roads fetch lower rents. These location characteristics explain 6 percent of the variation. Together, in column 7, they explain 17 percent of the variation.

Much of the remaining variation in the rental rates is at the village level. To illustrate this point, village-fixed effects are added. We note a jump in the R-square to 0.3 in column 8, suggesting village-level forces beyond plot-specific characteristics are likely to play a large role. If this is the case, norms could also prevent markets from correctly pricing information about field or soil characteristics within villages.²⁹

Insights from focus group discussions and key informant interviews appear to confirm that this village-level variation in rental prices may be related to social norms.

Interviews revealed that most farmers (both tenants and landlords) believe that it is unfair to decrease rental rates when supply exceeds demand for land or to increase rental rates when demand exceeds supply. Even when prompted, respondents noted the lack of response in rental rate to factors such as plot location and plot fertility. However, when a tenant cultivates tobacco (the main cash crop in the area), or a tenant is foreign to the village, it was considered only “fair” to increase the rate, as “that individual is rich.” It is worth noting that such a comment implicitly assumes that rental rates can be used to redistribute surplus away from relatively wealthy farmers. Correspondingly, when a tenant is local, cultivates maize, or is a friend, respondents noted the need for a rental rate that follows what is “usually done.” While never stated explicitly, we suspect that what is “usually done” is driven in part by a local fairness norm.

If rates are unable to adjust upwards under land pressure, owners of good-quality fields might prefer to cultivate those fields themselves instead of renting them out. Using the results of the soil tests, one can compare the quality of rented-in and owner-cultivated fields. The average active soil carbon content of an owner-cultivated field is 431, while that of a rented-in field is 371; i.e., the soil carbon content of an owner-cultivated field is about 10 percent higher compared to that of a rented-in field (although these differences are not statistically significant, with a p-value of 0.168).

The data contain additional measures of soil quality from the survey for a larger sample of fields. Using these data, table 5 regresses a binary indicator of whether a field is rented in on a set of subjective farmer-reported soil-quality variables. The results suggest important differences between rented-in and owner-cultivated fields. In particular, soils of rented-in fields are less likely to have good fertility. As a caveat, note that these soil characteristics exhibit some correlation. For example, a fine textured, clay soil is more likely to exhibit water logging, as water does not penetrate it easily. Similarly, coarse, sandy soil may be more prone to erosion. The correlation matrix of these soil characteristics is reported in supplementary online appendix table S1.5. For example, the correlation between good soil fertility and nutrient depletion is −0.30. The correlation between a nutrient-depleted soil and erosion is 0.38. To address this concern, we follow standards in soil science that distinguish between soil’s physical and chemical properties (Berazneva et al. 2018) and select one chemical property and one physical property for the regressions in table 5.

4.2. Survey-Measured Equity Preferences

Supplementary online appendix table S1.1 reports the results of the hypothetical survey questions aiming to capture equity preferences as they relate to landlords and tenants. While 863 households participated in the land-rental markets as tenants only, 449 did so as landlords. Column 1 includes all respondents, while columns 2 and 3 report the answers of the tenants, and landlords, respectively, following the same classification as in supplementary online appendix table S1.2.

One can draw three primary conclusions. First, most respondents are concerned with the welfare of tenants and value the relationships between tenants and landlords: 78 percent note it is (very) unfair to increase the rent to one’s regular tenant (scenario 1A), and 90 percent note it is (very) unfair to increase the rent of a tenant in need even though land is scarce (scenario 9B). Many respondents (65 percent) appear to be “against” the market clearing mechanism, i.e., an increase in rent when land is scarce (scenario 8). Even more, 73 percent, are against a Becker–DeGroot–Marschak (BDM)-style elicitation of the willingness to pay of tenants (scenario 6).

Second, while respondents express concern for the welfare of landlords, there appears to be an asymmetry in their concern (as compared to their concern for tenants). Nearly 60 percent of respondents were against raising the rents in a case where a new buyer was pushing up demand (scenario 4). However, when respondents were asked to imagine a reversal of the scenario in which market prices declined (scenario 5), only 30 percent responded that a corresponding decrease in rental rates would be (very) unfair. Recall that the survey was not set up to provide a complete comparison between concerns for landlords and concerns for tenants, hence these conclusions should be viewed as suggestive.

Third, comparing the answers from tenants and landlords, one notes few meaningful differences in their responses. If the observed norms are mostly preference driven, this would be expected. It is as if the village as a whole agrees on what is right and wrong, independent of who they are. While for 6 out of 12 questions there is a statistically significant difference between their responses, the magnitude of these differences is limited, ranging from a few percentage points to about 10. However, it should be noted that tenants appear to consider most scenarios (including those that relate to protecting landlords) more unfair compared to landlords. For example, when inquiring about market clearing-mechanisms in scenario 8, “Last year the rental rate was 15,000 kwacha. This year there are more tenants seeking to rent land. One landlord decides to charge an increased rate, 17,000 kwacha,” 69 percent of tenant households indicated this tobe unfair, while only 60 percent of landlord households thought this was unfair.

Supplementary online appendix table S1.6 presents a correlation matrix for all the inequity-averse preferences scenarios in supplementary online appendix table S1.1. One notes meaningful correlations between these variables. For instance, there are significant correlations between inequity-averse preferences related to social determinants of the rental rates and rental-rate response to shocks.

4.3. Experimental Measure of Fairness Norms: A Modified Dictator Game

While the survey measures document the presence of equity preferences, a norm is often thought of as a group-level force which affects behavior. As such, we set up the dictator game to measure collective behavior at the village level in response to a scenario that gives rise to between-group inequality. To summarize the discussion below, one observes a preference for Pareto-dominated, but equal, distributions of payoffs (Bundle A) in many villages. There is also a surprising degree of unanimity in the votes between group members, and even within villages as a whole, suggestive of the existence of a norm.

Table 1 panel 1 presents the results. All individuals across groups within the village are pooled to create village-level measures. On average, 38 percent of participants vote for Bundle A in the first round, 63 percent of participants vote for Bundle A in the second round. One notes that substantially fewer groups opt for equal distribution Bundle A in the first round compared to the second round. This suggests that inequality is especially disliked when one is behind others (consistent with Güth, Schmittberger, and Schwarze 1982, Fershtman, Gneezy, and List 2012, Charness and Rabin 2002).

When the game randomized which group receives the larger amount in Bundle B in Round 3 (and, recall, the result of this randomization is unknown at the time of the decision), 43 percent of groups opted for Bundle A. The Round 3 elicitation technique corresponds to the axiomatic concept of anonymity within any measure of inequality, as the identity of those who will receive the lower amount is unknown (Dasgupta, Sen, and Starrett 1973). As discussed below, the Round-3-based measure is also significantly correlated with the various measures of inequity aversion. As such, the regression analysis focuses on the third round for the main results, using the percentage of participants that vote for the equal distribution (i.e., Bundle A) in the third round as a measure of the strength of the fairness norm in the village.

One might be concerned that the result of this game could be affected by the sample composition: tenants versus landlords. If landlords tend to rent to more than one tenant, there are likely to be more tenants than landlords in the village, and hence in the sample (which appears to be the case—see table 3 and supplementary online appendix table S1.2). However, with few meaningful differences in preferences (see supplementary online appendix table S1.1) and a very fluid definition of who is a tenant and who is a landlord (see the discussion under descriptive analysis), the exact sample in both game and survey may be less important. To check this proposition, supplementary online appendix table S1.4 regresses the strength of the norm on the ratio of the share of landlords to the share of tenants in the village. The results indicate no statistically significant relationship, which reassures us that the elicited norms do not depend on sample composition. This suggests that what one is capturing is a village-level norm indeed.

Figure 3 presents the histogram of the percentage of the participants in each village who voted for the equal distribution (i.e., Bundle A) in Round 3. The distribution has three focal points at 0, 50, and 100 percent, with the largest number of villages at 50 percent. Zero percent means that no one in the village voted for the equal distribution, while 100 percent indicates that all voted for the equal distribution. One notes, for instance, that, in 6 percent of the villages, everyone voted for the “fair” bundle, while in 8 percent of the village, everyone voted for the “efficient” bundle—in the rest of the villages, there are disagreements (note that the bulk of these disagreements is between groups and not within groups). The largest mass is at 50 percent, seemingly suggesting disagreement within the village. However, one should keep in mind that each village had two groups, and if the two groups vote distinctly, but each in a unanimous manner, the resulting measure would also be 50 percent. In effect, when analyzing the votes at the group level, rather than the village level, as in supplementary online appendix fig. S1.4, one can see that the mass at 50 percent disappears, and indeed most groups’ vote was unanimous, meaning the group’s participants appeared to have come to a consensus prior to the vote. One notes that 43.6 percent of the groups made unanimous decisions (i.e., all the members in the group voted for either Bundle A or Bundle B). The unanimous nature of the votes is not unexpected given that the groups were encouraged to discuss their choices prior to each vote.

Supplementary online appendix fig. S1.5 presents the results for participants that voted for equal distribution (i.e., in Bundle A) in Rounds 1 and 2. The mass of the distribution shifts towards the left in Round 1 (where one would receive 1,000 MK in case of the unequal distribution) and towards the right in Round 2 (where one would receive 500 MK in case of the unequal distribution).

To conclude this section, supplementary online appendix table S1.6 presents the correlations between the modified dictator game results (i.e., the strength of the norm) and the survey-elicited preferences. The strength of the fairness norms correlate positively with the shares of households that answered (very) unfair to 8 out of 12 questions of the survey-elicited preferences.

5. The Role of Fairness Norms in Rental Rates and Market Participation

This section documents the role of village fairness norms in the land market. We map up village-level rental-rate measures with the strength of the village’s fairness norm, controlling for potential village-level confounders:

where ln RentRateMeasure_v is the natural log of per acre rental-rate range/minimum/maximum for village v, and |$\it{SFairnessNorm}$| measures the strength of fairness norm in village v. The parameter β is the coefficient of interest. The variable |$\it{SFairnessNorm}$| is measured as the percentage of participants in village v who chose the equal income distribution in Round 3 of the dictator game. The vector |$\boldsymbol{X}$| is a set of village-level control variables capturing market access, population pressures, and migration. The selection of these variables was guided by literature on households’ participation in the land-rental market in Malawi (Chamberlin and Ricker-Gilbert 2016; Abay, Chamberlin, and Berhane 2021).

Table 6 reports results from the estimations. Results reported in column 1 do not include village controls; column 2 includes village controls. Results are comparable across columns; the discussion focuses on the results reported in column 2.

The findings in column 2 indicate that a 1 percentage point increase in the strength of fairness norms is more likely to be associated with a 0.325 percent reduction in a village’s rental-rate range. Extrapolating this number, in a village where 40 percent of the households vote in favor of the fair allocation, the village’s rental-rate range is 13 percent (i.e., 40*0.325) more likely to be lower than what it would have been in the absence of any fairness norm.

Whether the norm is associated with a relative increase in the minimum rental rate or a relative reduction in the maximum has implications: Are norms serving to protect landlords or tenants, or both? Columns 3, 4, 5, and 6 of table 6 document the relationship between the strength of fairness norms and the village’s minimum and maximum rental rates. The results indicate that fairness norms mostly act as a means of protecting tenants: the stronger the fairness norm, the lower the village’s maximum rental rate. The results show that a 1 percentage point increase in the strength of fairness norms (i.e., column 6) is more likely to be associated with a 0.23 percent reduction in the village’s maximum rental rate. The result suggests that if 40 percent of households vote in favor of the fair allocation, then the village’s maximum rental rate is 9.2 percent (i.e., 40*0.23) more likely to be lower than it would have been in the absence of any fairness norms. One finds no significant evidence of norms protecting landlords.

While the descriptive statistics on inequity preferences suggested the presence of concern for poorer landlords, results here indicate that overall the norms, as measured by the dictator game, may not adequately be protecting poor landlords by failing to pull the minimum rent up. In this context, landlords and tenants have equivalent wealth and household size (see supplementary online appendix table S1.2). This is unlike other studies set in the area; see, among others, Ricker-Gilbert et al. (2019) which notes that tenants appear to be wealthier than their landlords.

Supplementary online appendix table S1.7 reports the results using alternative outcome measures: the log of rental-rate variance, absolute skewness, and the ratio of the range and median rental rates. Consistent with the results reported in table 6, fairness norms are associated with a lower village rental-rate variance: a 1 percentage point increase in the strength of the fairness norm is more likely to be associated with a reduction in the rental-rate variance by 0.641 percent.

A last robustness check is a split sample analysis, dividing the village sample into villages with a large number of tenants and villages with a lower number of tenants. A “high-percentage tenant village” is a village with a share of tenants greater than the median across all villages, while a “low-percentage tenant village” is a village with a share of tenants less than or equal to the median across all villages. The results are reported in supplementary online appendix table S1.8. As in supplementary online appendix table S1.4, one again finds that the exact composition of the village, in terms of landlords versus tenants, does not matter much. This might be due to the fluidity of the definition, as people move in and out of landlord/tenant status, or due to the fact that preferences among these two groups, as stated earlier, are quite comparable overall.

Fairness norms might also impact the amount of land being rented in and out, and could also prevent markets from adjusting to other market considerations. Prevailing rental rates, for example, might not reflect the agronomic quality of fields. As rates are unable to adjust upwards, landlords of good-quality fields might prefer to cultivate those fields themselves instead of renting them out. The last step is to examine the relationship between fairness norms and market participation by replacing the outcome variable in equation (1) with variables such as the total acreage of rented-in land divided by the total acreage owned, the percentage of households in the village that participated in the market in 2018 (as either tenant or landlord), the percentage of rented-in fields that are eroded, that are of good fertility and with fine-textured soil. The results of these estimations are reported in tables 7 and 8.

Table 7 presents the relationship between rental-market activity and the fairness norm. Column 1 considers the share of rented-in land (acreage, as the ratio of total land owned) referring to the 2017–18 growing season. Column 2 analyzes the number of households that participated in the land-rental market in the 2017–18 growing season. Columns 3 and 4 study the number of households who have participated in the land-rental market in the past 10 years as tenants and as landlords. One finds no statistically significant relationship in columns 1 and 2 respectively, which assess participation in land-rental markets in only the 2017–18 growing season. When considering households’ long-term participation in the market in columns 3 and 4, there is no statistically significant correlation between the strength of the norm and both tenant and household participation in the land market.

Table 8 links measures of soil fertility to the fairness norms. The dependent variables are the village-level percentage of rented-in fields that are (a) eroded, (b) with good fertility, (c) with fine textured soil, and (d) with any form of degradation.³⁰ While table 5 reported that rented-in fields were of poorer quality, one now notes no evidence of a statistically significant relationship between the norms and these measures of soil quality.

6. Conclusion

Cultural inheritance, including the social norms and practices influencing decision-making in communities across the world, plays an important role in shaping economic outcomes (Michalopoulos and Xue 2021; Jang and Lynham 2015). This paper studies how prevailing social norms, and in particular norms related to fairness, relate to the functioning of land-rental markets in rural Malawi.

The paper documents the relevance of social norms to land-rental markets across Central Malawi using primary data from farmer surveys and a dictator game, the latter modified to elicit group preferences for equal distribution of resources. This experimentally elicited measure suggests the presence of strong norms oriented towards fair distributions over Pareto superior, but unequal, distributions.

Linking these norms to land markets, the results show that the observed range of the per acre land-rental price contracts as the experimentally measured fairness norm increases, even after controlling for potentially confounding factors including population pressure, market access, and migration. While fairness norms for both tenants and landlords appear present, the results indicate that the fairness norms for tenants appear to be quantitatively more important. The land-rental price range tends to be constrained by a price ceiling rather than a price floor. The results of this study suggest several new areas of research.

First, this analysis is exploratory and descriptive, documenting the presence of inequity preferences and fairness norms through survey data and games, and providing a first analysis of their potential relevance to Malawi’s land-rental markets. This study does not investigate the origins or historical evolution of these norms, the subject of substantive contributions in evolutionary biology, game theory, and social psychology (Young 2015; Henrich 2017; Henrich and Muthukrishna 2021). For example, Guerriero (2020) notes a relation between climate, terrain, and agricultural conditions and today’s norms of cooperation in Europe.

In the Malawian context, social customs including patrilineal and matrilineal inheritance rules long dictated who inherits the land, and village chiefs acted as arbitrators (Takane 2008), although increased population pressures on land, increased male mortality due to HIV/AIDS, and increased rural-to-rural migration have contributed to some degree of flexibility. As such, further investigations on the origins of the fairness norms might begin by documenting existing practices of sharing and distribution (as in Jakiela 2015), the villages’ characteristics (as in Iacobelli and Singh 2020), and movement/migration in and out of the villages (as in Fafchamps and Hill 2019; Luke and Munshi 2011).³¹

Second, connections exist between fairness norms and other economic processes in low-income countries. The presence and implications of incomplete and imperfect markets have been a central feature of economists’ understanding of behavior, poverty, and inequality in low-income countries (Stiglitz 1988). Social structures can alter opportunities and enhance productivity by providing insurance, credit, and information (Fafchamps 2011; Hong and Kacperczyk 2009). However, these structures can also constrain (Munshi 2014). For instance, sharing norms within extended family networks in Africa have been shown to limit investment in education and business (Di Falco and Bulte 2015; Cox and Fafchamps 2007). If social norms constitute an additional constraint, how can one expect their role to change when other constraints are relaxed or lifted, for instance, when insurance markets expand, or the technologies introduced are progressive in nature?

From a policy perspective, Malawi’s Customary Land Act³² fosters registration of customary land as private land, to boost productivity through increased investment and increased participation in land-rental markets. This formalization has increased land sales and rentals (Chamberlin and Ricker-Gilbert 2016) and has driven a transformation from land borrowing between relatives to seasonal and formal rental agreements mostly between non-relatives. The results in this paper call into question whether such reforms will result in the anticipated benefits not only in terms of efficiency but also in terms of distribution. For example, reforms might increase social conflict between “original settlers” and “strangers” if distributional concerns are not adequately addressed (Peters and Kambewa 2007; Takane 2008).

Finally, it may be possible for policy experiments to consider nuanced and varied institutional designs that emerge from processes of action research (Pande and Udry 2005; Bowles 2016). Processes in which a feedback loop is created between government and local actors so that policy changes can improve the functioning of land markets in a manner consistent with locally desired outcomes in an adaptive and incremental fashion (Andrews, Woolcock, and Pritchett 2017) are particularly of interest.

Declaration of interest statement

None.

Data Availability Statement

Data used in this study are primary data collected by the authors. Anyone interested in the data may contact the corresponding author. Stata .do files for replication will be made available on the corresponding author’s website (https://sites.google.com/view/kwabenakrah/home).

Acknowledgement

The authors acknowledge financial support from the International Initiative for Impact Evaluation (3IE) under TW4.1018.

Author Biography

Kwabena Krah (corresponding author) is a postdoctoral fellow at the Institute for Racial Justice at Loyola University Chicago, United States; his email address is [email protected]. Annemie Maertens is a professor in the Department of Economics at the University of Sussex, United Kingdom; her email address is [email protected]. Wezi Mhango is an associate professor of Agronomy at Lilongwe University of Agriculture and Natural Resources, Malawi; her email address is [email protected]. Hope Michelson is an associate professor in the Department of Agricultural and Consumer Economics at University of Illinois Urbana-Champaign (UIUC), United States; her email address is [email protected]. Vesall Nourani is a director and lead researcher at the Development Innovation Lab and a senior research associate in the Kenneth C. Griffin Department of Economics at the University of Chicago, United States; his email address is [email protected]. The authors acknowledge financial support from the International Initiative for Impact Evaluation (3IE) under TW4.1018. The authors are grateful to Nathan Nunn, Alex Winter-Nelson, Kathy Baylis, Ben Crost, Matt Lowe, Alan De Brauw, Laura Schechter, Adriana Kugler, Rachel Heath, S Anukriti, Shing-Yi Wang, Melanie Khamis, and seminar participants at the CSAE 2019, AAEA meetings 2019, NEUDC 2019, and the Vancouver School of Economics, and two anonymous referees for comments and/or discussions. The authors also thank the Wadonda Consulting field team for data collection and the late Ephraim Chirwa for supervising the data collection. Project coordination was provided by Eric Kaima, and additional research support by Annie Matiti. A supplementary online appendix is available with this article at The World Bank Economic Review website.

Footnotes

References