Intervention(s)
The interventions in this experiment involves measurement and elicitation of value of rural assets, including land, livestock, productive, and durable assets. Each sampled household's asset roster includes a fixed list of durable and productive assets (including land, livestock, agricultural tools, furniture, and cooking/storage equipment). Each household is randomly assigned to exactly one of three elicitation and valuation-question framings, used consistently across every asset in its roster: (T1) a self-reported sale-value "sell" frame ("If you were to sell this [ASSET] in the same condition today, how much SDG would you expect to receive?"); (T2) a self-reported replacement-cost "buy" frame ("If you were to buy this [ASSET] in the same condition today, how much would you expect to pay (in SDG)?"); and (T3) a neighbor-referenced “sell” frame ("If your neighbor was to sell this [ASSET] in the same condition today, how much SDG do you expect he would receive?"). Independently of arm assignment, for the full sample, enumerators will separately collect local market prices for the same asset categories from vendors/markets serving the sampled communities (T4), providing a common benchmark against which all three self-report arms can be compared.
Implementing these measurement methods in the same setting and among large sets of assets generates a number of unique contributions. First, despite the widespread use of these measurement and elicitation methods in several studies, we are not aware how these methods are comparable and if they generate higher or lower valuation of assets. Second, eliciting these among various types of assets will identify vulnerability and limitations of these methods across different asset types and groups.
Below we describe each of the measurement methods in more detail:
T1 — Self-reported sale value
This method elicits what respondents would receive if they were to sell the specific asset in the same condition today. This is the modal approach in LSMS-type multi-topic household surveys: respondents are asked how much money they would receive if they were to sell an asset, and these values are aggregated across assets to construct wealth aggregates (Carletto et al., 2013). The approach is attractive and less costly because it does not require market survey, but it suffers from two limitations. First, it is a hypothetical valuation, and hypothetical elicitation is known to generate "hypothetical bias" because respondents have weak incentives to answer carefully. Similarly, such valuation exercises suffer from social desirability bias, a concern that arises when valuations are influenced by what respondents believe is an appropriate or expected answer rather than their true assessment. Second, some respondents might not have current market information to accurately assess the value of a specific asset. Self-reported values of durable, illiquid assets specifically have been shown to be unreliable even in contexts with reasonably developed markets (e.g., Gonzalez-Navarro and Quintana-Domeque, 2009). This motivates treating T1 not as a gold standard but as one candidate elicitation frame to be benchmarked.
T2 — Self-reported replacement cost
This approach elicits what it costs to buy an asset in the same condition today. This differs from T1 only in the direction of the hypothetical transaction — buying versus selling the same item in the same condition — which makes it a direct test of the endowment effect, reflected through the willingness-to-accept–willingness-to-pay (WTA-WTP) gap documented in behavioral economics literature (e.g., Kahneman et al., 1990). In many settings, the compensation participants demanded to give up an owned item ("willingness to accept") was roughly twice what other participants were willing to pay to acquire the identical item, a finding interpreted as evidence that losses are valued more heavily than equivalent gains (Kahneman et al., 1990). Implementing both a "sell" (T1) and a "buy" (T2) frame for the same physical assets in a rural household context enables us to test whether the WTA–WTP gap documented mostly in lab and stated-preference settings also shows up in ordinary household reporting of durable goods, which has direct consequences for asset-based wealth construction. If households systematically report higher replacement-cost values than sale values, standard modules that only ask one framing will produce mechanically different wealth rankings depending on which question was used. This has important implications for informing the design of rural household surveys.
T3 — Neighbor-referenced valuation
This elicitation method aims to test whether some of the bias in T1/T2 is driven by strategic misreporting or social-desirability concerns tied to the respondent's own asset holdings and related access to markets. For example, respondents might want to strategically underestimate the value of their assets for many reasons— a concern that is especially salient in contexts, like humanitarian and social-protection targeting surveys, where households may believe reported wealth affects program eligibility. Household survey data are known to suffer systematic measurement error driven by recall bias, strategic misreporting, and social desirability bias, and this has motivated a push toward direct measurement tools precisely because self-reports are compromised by respondents' incentives. Related work on socially sensitive valuation finds that growing evidence points to social desirability bias distorting stated valuations because respondents answer in ways they believe match the interviewer's expectations rather than their true assessment, and the broader survey-methods literature confirms that no single question-design fix eliminates social desirability bias, but indirect and third-person framings are among the standard mitigation strategies used to reduce it. Shifting the referent from "you" to "your neighbor" is a form of indirect questioning: it keeps the valuation task (price a given asset in given condition) identical to T1 while removing the respondent's own stake in the answer. This can attenuate both strategic under- or over-statement since the respondent is not being asked to part with or acquire anything themselves.
T4 — Enumerator-collected local market price
This method aims to tackle respondent bias by avoiding respondent self-report altogether: enumerators independently collect prevailing prices for the same asset categories in local markets. This follows the logic used to construct unit values and price data for consumption aggregates in household surveys. This mirrors recent measurement work on other asset categories in African household surveys, where direct, non-self-reported measurement has been shown to correct substantial bias relative to standard respondent-based modules: direct enumerator counting of livestock, for instance, increased total reported livestock ownership by 39 percent and cattle counts by 43 percent relative to self-reported baseline data (Abay et al., 2025). This method is intended to function as the closest available approximation to an objective, incentive-free benchmark against which T1–T3 can be evaluated, while acknowledging that market prices themselves are imperfect proxies for a specific used asset's condition-adjusted value — which is precisely why the experiment also varies the self-report framing rather than assuming any one method is correct by construction. Although market prices collected by enumerators might also suffer from some important limitations, we hypothesize such biases are less likely to be systematic and related to respondent characteristics.
Research questions:
This experimental study aims to address the following research questions:
(i) Do self-reported household asset values differ depending on whether the valuation question is framed as a hypothetical purchase (replacement cost) versus a hypothetical sale?
(ii) Does shifting the reference of the valuation question from the respondent's own asset to a hypothetical neighbor's sale change the value and variance of reported values?
(iii) How well do self-reported asset valuations (under each of the three framings) approximate an independent, enumerator-collected local market price for the same asset category? And:
a. How systematic are these differences?
b. Do these vary across the three self-report frames?
c. Do these vary across asset types and groups?
(iv) Does the choice of asset valuation have material consequences for downstream applications, such as asset-based wealth ranking and classification?