{"template":"article","title":"Retention of Title and Demand for Irreversible Disposal","authors":[{"name":"Maëlle R. Dervaux","affiliations":[1],"email":"m.dervaux@varenne-economics.example","corresponding":true},{"name":"Azra D. Ertekin","affiliations":[2]},{"name":"Jonas A. Verheijen","affiliations":[3]}],"affiliations":["Department of Applied Economics, Institut Varenne d'Économie, Lille, France","Centre for Household Finance, Alderbroek Institute, Utrecht, Netherlands","Department of Economics, Institut für Grenzmarktforschung, Aachen, Germany"],"abstract":"Providers of irreversible disposal services can act as agents of the owner or acquire title as part of the transaction. In a posted-price experiment with 640 households at four sites, we vary the disposition of title while holding the service provider, payment schedule, and physical outcome fixed. Title transfer increases acceptance by 15.3 percentage points over five signed offers (95% confidence interval 8.4 to 22.2). At zero compensation, 51.6% accept under transfer, compared with 28.1% under agency. Both contracts eliminate the owner's right to recover the object at the same moment. The data identify the effect of offering a different contract at each price. An interpretation in euros requires assumptions about the reservation amount between the prices we observe. We report bounds for the mean reservation amount over the sampled price range and discuss what the contract comparison can establish about the value of retaining an object with an unverified use.","keywords":["property rights","household durables","disposal","posted prices","option value"],"body":"# Introduction {#sec:intro}\n\nA service provider can remove a durable good under a mandate from its owner or take it into ownership before removing it. Where the provider has discretion over subsequent use, the two contracts allocate a residual right with an evident economic value. Where the good is to be destroyed under either contract, the distinction is less clear. A sufficient comparison needs to hold fixed the point at which the former possessor relinquishes control as well as the ultimate destination of the good. Otherwise a preference for one document can be a preference for a few more days in which to change one's mind.\n\nWe arrange such a comparison in four local intake offices. The price is posted, either party can be on the paying side, and each household sees a single offer for one object it already owns. Acceptance authorizes the same irreversible disposition in both arms. Under agency, the provider destroys the object on the household's behalf and title extinguishes at destruction. Under transfer, title passes to the provider when the offer is accepted and extinguishes at destruction. In the latter case the provider has a contractual duty to dispose of the object in the same manner. We ask whether people respond differently when the object belongs to the provider while it is being disposed of.\n\nThe question arises in settings where removal costs exceed the scrap value of the good and specialist handling costs exceed the price of the service [@loret2018; @sauer2022]. Most records available from providers begin after an owner has decided to part with the object. A difference in the share of goods discarded under different contracts in such records could reflect selection into the intake office, local prices, or the provider's choice of which contract to offer. We draw both the contract and the price after intake. The offer therefore addresses a household at the last point where it can take the object home.\n\nThe object is a single metal key from the household's own stock. We restrict intake to keys for which the owner can identify no current lock and no owner outside the household. A participant who knows that the key belonged to a door that has been replaced is eligible on the same terms as a participant who cannot remember where the key came from. The distinction matters for valuation. Some keys have no remaining use. Other keys may still open something, and the only person who could say what they open may never think to ask for them. The eligibility screen leaves both possibilities open.\n\nThe common option of taking the key home lets us compare the acceptance schedules under agency and transfer without estimating the probability of a future match. We find a gap at every posted price. The gap is largest around zero compensation, where both contracts attract a substantial number of households. We use the grid in @sec:results to report the range of implied changes in clipped reservation amounts. @sec:discussion considers why passing title for the last part of a key's useful physical existence could affect the decision.\n\n# Ownership and the disposal decision {#sec:framework}\n\n## Control and the interval before destruction\n\nThe comparison concerns a narrow right. In the terminology of @barzel1997, legal title and the ability to derive value from an asset can be allocated differently. An authorized disposal agent has custody and a limited mandate. In our agency contract the owner chooses the terms of disposal at the point of decision and delegates all actions that implement those terms. Under transfer the provider has title and is bound by the same terms. Neither party has a right to redirect an accepted item into a resale channel. Payment is conditional on the service being completed, and the item returns to the household on identical terms if the provider cannot complete it. An owner always retains the right to leave with the object by declining.\n\nThe point bears on comparisons based on @coase1960. An arrangement that moves title may also change which bargains are feasible after agreement. That feature matters in markets for usable secondhand goods with privately observed quality [@akerlof1970]. Here we constrain those subsequent bargains by agreement and keep the physical disposition auditable. Case studies of household clearances [@sevrin2019] describe a separate practice in which a buyer takes the whole lot and can sell items recovered during sorting. Lot buyers and removal agents offer different services in that setting, even when their original quotes match.\n\n## Keeping the object\n\nLet $K_i$ denote the money equivalent of retention, net of storage and future search costs, for household $i$. If a future cue can identify a use for a key, retention has an option value even when the present use is nil. This is the usual point about irreversibility [@arrow1974; @dixit1994]. Its size here is anyone's guess. A replacement key could cost very little if the lock were known. If the only known copy opens an inaccessible lock, the alternative to using it could be replacing the lock. We collect no lock information at intake, and people typically cannot attach a replacement price to an object whose function they cannot name. @neve2020 finds that households seldom record a possible use for an unidentified spare as a purchase they might otherwise have to make. Household inventory values based on observed replacement expenditure can omit precisely this class of use.\n\nWrite $d_{iz}$ for the money-equivalent cost of authorizing a disposal contract $z$, where $z=A$ is agency and $z=T$ is transfer. The cost may depend on how the disposition is described or on how the household understands the allocation of responsibilities. We use a local quasi-linear representation over the price range. At signed payment $p$ to the household, acceptance is\n\n$$\nY_i(z,p)=\\mathbf{1}\\{p \\ge R_{iz}\\},\\qquad R_{iz}=K_i+d_{iz}. \\label{eq:threshold}\n$$\n\nFor a household with $R_{iz}<0$, some positive fee for completing the contract is acceptable. The retention value can be negative if the household would prefer to clear the object from the stock. In that case the owner could still decline agency if its authorization cost is positive and sufficiently large. An alternative preference representation with income effects would require that the net advantage of agreement increase with $p$. For the threshold calculations, that monotonicity suffices [@hanemann1984]. The main contract comparison uses observed choices at each offered price and needs neither a utility representation nor quasi-linearity.\n\nA model in which households care about payment and control rights alone, understand the offers as specified, and expect identical performance assigns equal costs to the two documents. Large retention values can depress both schedules while preserving that equality. Under @eq:threshold a common downward shift in the reservation amount under transfer would move acceptance to the left. A variable shift can yield a similar pair of aggregate curves. Since a household sees only one offer, the five treatment contrasts average over households that may respond in different directions. Even marginal distributions known at every price would leave the joint distribution of the two reservation amounts undetermined.\n\nThe disposal decision differs in two ways from the endowment experiments of @kahneman1990. All participants start as owners, and everyone who accepts relinquishes the same physical asset. The starting reference point [@thaler1980; @samuelson1988] plausibly accounts for some rejection in both arms. Under both offers the default in the absence of an affirmative decision is that the owner takes the key home. What changes is the form of the affirmative act. We kept the exchange short and the words on screen close to the signed agreement. As in @plott2005, what a participant takes an elicitation to mean matters even if the intended meaning is clear to the staff. The pilot described in @sec:implementation preceded the experiment for that reason.\n\n# Field experiment {#sec:design}\n\n## Recruitment and screening\n\nWe staffed intake offices in Lille, Ghent, Utrecht, and Aachen between March and June 2025. Notices advertised a paid study of decisions about small household goods and described the eligible item and the possibility that it would be destroyed. A booking could cover only one household. The person attending had to be an adult, have authority to decide on the disposition of the item, and bring one eligible key to intake. The notice specified a participation payment of €12 and a maximum possible charge of €4. A participant received the €12 after completing the decision screen, irrespective of choice. The offer could increase or reduce that payment. Staff had no discretion to vary either amount.\n\nOf 962 screening records, 243 failed eligibility. The first recorded reason was a known current use in 98 cases, identifying marks or an electronic component in 78 cases, and authority or consent in 67 cases. These counts use the screening order and are mutually exclusive. Of the remaining 719, 79 did not attend within the intake period. We reached the site quotas of 160 assignments each with the 640 remaining households. Staff recorded confirmation of eligibility and consent before drawing an allocation. If a participant brought several keys despite the instructions, staff asked the participant to nominate a single eligible one before seeing any price or contract. We accepted one nomination per household and did not take possession of the rest. The sample consists of households that could and would bring an eligible key to an office. Owners of all never-identified keys are a substantially broader population.\n\nThe baseline form was completed before the offer. It asked whether the household recalled the key's origin and whether anyone in the household had used it during the previous year. The questions concerned the nominated item. A separate code entered by the intake worker recorded eligibility and site. The Institut Varenne review committee approved the consent and intake procedure (IVE-24-19). Participants could decline enrollment and leave with the item before the offer was drawn.\n\n## Offers and implementation {#sec:implementation}\n\nWithin each site a concealed assignment list contains 16 copies of each of ten offer labels. The labels are the cross of two contracts with payments of −€4, −€2, €0, +€2, and +€4 to the household. The list is permuted uniformly once per site. The intake worker opens the next entry only after recording consent and enrollment. This is a site-blocked completely randomized design; the four offices have separate lists. The resulting price distribution is the same by site and contract. The design lets us discuss demand at zero below without adjusting for the composition of the remaining offers.\n\nEach screen presents the verb used in the agreement, the common service terms, two choice buttons, and the net amount to be settled. The text at zero states explicitly that no amount will be added to or deducted from the participation payment. Declining leaves the key and the €12 with the participant. An affirmative choice requires an on-screen confirmation in both arms. The decision is recorded at confirmation, or when the household declines. Written terms govern custody from confirmation to disposition; the household has surrendered the right to recover the key during that period under either contract. A provider failure voids the contract, returns the key, and cancels the signed payment. Full operative language appears in @sec:documents.\n\nA 24-household pilot, completed at a separate office, tested the translation and settlement screens. Its offers and choices are outside the analysis sample. We changed the display of negative payments after some pilot participants read a minus sign as a deduction from a displayed fee. The final screen gives the signed amount in words alongside the arithmetic. Piloting was complete before the site lists were drawn. The main sample saw no practice offers, repeated quotes, or examples using nominated items. The intake worker could read the text aloud on request and was instructed to repeat the printed words if asked who would own the key immediately after confirmation. The worker had no opportunity to offer the other agreement. Prior market experience can change how a subject understands a sale [@list2003]; here the providers offered no other services during the experiment.\n\nThe same provider served both assignments at a site. Cashier and disposal staff handled the same numbered containers, with no title information on the physical label. A site ledger linked a final disposition or void to the offer number. Two accepted items, one under each contract, were returned because the disposition service did not complete before the session closed. Both administrative voids occurred on booked afternoons when an intake office could accept a choice but could not finish the last processing batch. The household retained the full €12 and recovered the key. We retain the two agreements in the acceptance outcome. The completion outcome records each as an incomplete service. There were no missing decisions or losses between enrollment and offer. @tbl:participation gives the accounting.\n\nTable: Household decisions and delivery, by assigned contract. All 640 enrolled households received an offer. A void is an accepted agreement whose disposition could not be completed. {#tbl:participation}\n\n| Stage | Agency | Transfer | Total |\n|:--|--:|--:|--:|\n| Assigned and offered | 320 | 320 | 640 |\n| Declined | 212 | 163 | 375 |\n| Accepted | 108 | 157 | 265 |\n| Of accepted, service completed | 107 | 156 | 263 |\n| Of accepted, service voided | 1 | 1 | 2 |\n\n## Estimands and inference {#sec:estimands}\n\nThere are $S=4$ sites and $J=5$ prices. Let $\\bar Y_{sjz}$ be the mean in price cell $j$ of contract $z$ at site $s$. Our primary estimate compares acceptance under transfer and agency, weighting sites and prices equally as planned before the experiment.\n\n$$\n\\widehat\\Delta = \\frac{1}{SJ}\\sum_{s=1}^S\\sum_{j=1}^J\n(\\bar Y_{sjT}-\\bar Y_{sjA}). \\label{eq:diff}\n$$\n\nFor price-specific contrasts we average the corresponding differences across sites with weight $1/S$. The scientific target for @eq:diff is the sample-average contrast under a uniform draw from the five offers. A market with more mass at the endpoints would have a different acceptance effect. The uniform weighting is useful here because it gives the two sides of zero equal representation without assigning all weight to a fee or all weight to compensation. It is a fixed rule and does not reflect observed uptake.\n\nWe report design-based standard errors using the finite-population upper variance estimator for a linear contrast in a completely randomized multiarm experiment with site blocks, and use a normal critical value for intervals [@imbens2015]. The upper variance estimator omits the subtraction for the unobserved variance of unit-level contrasts. See @sec:variance for the expression. The exact test of the sharp null that contract makes no difference to any choice holds price assignments fixed and permutes contracts within site and price. Its reference distribution conditions on the observed number of acceptances in each such stratum. The acceptance difference in @eq:diff is the sole primary contrast. The site quotas give a worst-case standard error of about 4.1 percentage points for that contrast after omitting the subtraction term. The displayed intervals at individual prices are pointwise, and the site results are descriptive checks on the aggregation.\n\n# Acceptance and reservation amounts {#sec:results}\n\n## Response at posted prices\n\nThe overall difference is 49 agreements in two equally sized groups. Transfer raises acceptance from 33.75% to 49.06%, a difference of 15.31 percentage points (95% CI 8.40 to 22.23). The exact two-sided test gives $p=0.0000202$ under the sharp null. Completed dispositions rise by the same number of agreements. Because only one agreement was voided in each arm, reclassifying both failures as zero yields the identical point estimate with a different variance.\n\nTable: Uptake at each offered payment to the household. Each cell rate has denominator 64. The final two columns are in percentage points. {#tbl:prices}\n\n| Payment (€) | Agency, accepted | Transfer, accepted | Difference | 95% CI |\n|--:|--:|--:|--:|:--|\n| −4 | 6 | 10 | 6.25 | −5.48 to 17.98 |\n| −2 | 11 | 20 | 14.06 | −1.00 to 29.13 |\n| 0 | 18 | 33 | 23.44 | 6.46 to 40.42 |\n| +2 | 30 | 43 | 20.31 | 3.01 to 37.62 |\n| +4 | 43 | 51 | 12.50 | −3.12 to 28.12 |\n| Pooled | 108 | 157 | 15.31 | 8.40 to 22.23 |\n\nAs @fig:demand shows, even with €4 compensation, 21 of 64 households offered agency and 13 of 64 offered transfer keep their key. At the other end, 16 of the 128 households quoted a €4 fee accept one of the two services. We observe substantial uptake when an owner pays and substantial retention when the provider pays. Neither extreme price produces evidence that a common boundary value would describe almost all households. The range is wide enough to bracket a median under each contract in the point estimates and too narrow to locate the mean reservation amount without a tail assumption.\n\n```chart {#fig:demand caption=\"Share accepting at the assigned payment to the household. Marks show the four-site means and bars show pointwise 95% intervals from the upper design variance estimator. Lines connect observed prices.\"}\n{\n  \"type\":\"line\", \"xlabel\":\"Payment to household (€)\", \"ylabel\":\"Acceptance (%)\", \"x\":[-4,-2,0,2,4],\n  \"series\":[\n    {\"label\":\"Agency\", \"y\":[9.375,17.1875,28.125,46.875,67.1875], \"err\":[7.33,9.52,11.35,12.60,11.86], \"marker\":\"s\", \"style\":\"dashed\"},\n    {\"label\":\"Transfer\", \"y\":[15.625,31.25,51.5625,67.1875,79.6875], \"err\":[9.15,11.68,12.63,11.86,10.16], \"marker\":\"o\"}\n  ], \"legend\":\"upper left\", \"ylim\":[0,100], \"size\":[5.2,3.15], \"style\":\"pgfplots\"\n}\n```\n\nAcross sites the estimated differences vary between 15 and 16.25 points (@fig:sites). The precision of a site-level contrast is much lower than that of the pooled estimate. The observed similarity is compatible with some heterogeneity in the underlying effects. In particular, we have one translation and one office per site. The estimate does not isolate the effect of the word used for title in any one language. The agreement terms and instructions were reviewed jointly before translation. An independent reader of each back-translation checked the timing clauses before intake opened. We may still be seeing differences in the ordinary-language pragmatics of being asked to act as seller versus principal.\n\n![Within-site effect of transfer on acceptance, in percentage points. Bars are marginal 95% design-based intervals. The horizontal line marks the pooled point estimate.](site-contrasts){#fig:sites}\n\nThe gap of 23.44 points at zero combines two acceptance decisions with no signed settlement, during the same intake visit. This comparison speaks to an explanation in which all of the contract effect comes from the discomfort of paying to get rid of a possession. At zero both groups have the same participation payment regardless of choice. The words that accompany payment still differ across contracts, and the zero cell cannot separate the content of the authorization from what a household infers from those words. There is also little power to decide whether the gap peaks at zero, at a small positive amount, or over a flat interval that includes both. The center of the grid is where both curves have room to move.\n\n## Translation to euros {#sec:money}\n\nAn owner with a reservation amount between adjacent grid points accepts only the higher of the two prices. The design estimates fractions of such owners without locating them inside each interval. Consider the restricted monetary summary $\\mu_z$ formed by clipping $R_{iz}$ to the price range and averaging across households. Under @eq:threshold, if $F_z$ is the distribution function of the reservation amount in the study population, then\n\n$$\n\\mu_z = \\mathbb{E}[\\max(-4,\\min(4,R_{iz}))]\n       = 4-\\int_{-4}^4 F_z(p)\\,dp. \\label{eq:clip}\n$$\n\nAt any price in an open grid interval, the corresponding left and right endpoint probabilities bound $F_z(p)$. As in other bounds approaches [@manski2003], this contains an identification step and a sampling step. Inserting the observed rates into the two endpoint sums gives €0.81 to €1.97 under agency and −€0.59 to €0.69 under transfer (@fig:bounds). These are estimated ranges for the restricted means. They permit a reduction in the restricted mean of €0.125 to €2.563. The normal outer confidence interval, constructed by taking a one-sided 97.5% confidence limit at each end, extends from −€0.52 to €3.19. The results establish the acceptance contrast more precisely than a lower bound on the mean amount saved.\n\n```chart {#fig:bounds caption=\"Restricted mean reservation amounts between −€4 and +€4. Vertical segments mark plug-in identification bounds over all nondecreasing within-grid curves. Squares mark the piecewise-linear estimates.\"}\n{\n  \"type\":\"scatter\", \"xlabel\":\"Contract (1 = agency, 2 = transfer)\", \"ylabel\":\"Restricted mean (€)\", \"x\":[1,2],\n  \"series\":[{\"label\":\"Bounds\", \"y\":[1.390625,0.046875], \"err\":[0.578125,0.640625], \"marker\":\"s\"}],\n  \"hlines\":[{\"y\":0,\"style\":\"dotted\"}], \"legend\":\"none\", \"xlim\":[0.5,2.5], \"ylim\":[-1,2.3], \"size\":[4.6,2.8], \"style\":\"pgfplots\"\n}\n```\n\nLinear interpolation of the five probabilities yields a restricted mean of €1.39 for agency and €0.05 for transfer, a difference of €1.34 (conditional on the interpolation, 95% CI €0.74 to €1.95). The curve under agency crosses 0.5 between €2 and €4, while the transfer curve crosses 0.5 between −€2 and €0. Linear interpolation puts the two crossings at €2.31 and −€0.15. The difference between those crossings is a difference of marginal medians. It need not equal the median difference an owner would report when offered both arrangements. It also differs from the mean change in @eq:clip, which weights all four gaps between prices and limits the influence of households outside the grid.\n\n# Discussion {#sec:discussion}\n\nThe service is unusual in that the provider acquires title only after the household has made the decision to dispose. At that point title comes with no opportunity to choose a use. The office processed both contracts in the same disposal batches. Their administrative records identify the household or the office as the last owner. Under agency, the office has recorded that it destroyed somebody's key. Under transfer, it has recorded that it destroyed its own.\n\nSeveral readings of that clause fit the response. A household may treat a sale, including a negative-price sale, as a more familiar transaction than a mandate to destroy personal property [@loret2018]. The new owner may be seen as assuming some responsibility for the decision even when its action is specified in advance. An implied evaluation by the office could enter as information: a person may suppose the office would choose to buy only a key for which a useful match is unlikely. Assignment was random and the printed terms state that the provider follows the same disposition for all accepted items. Either fact might carry less weight than the household's experience of what it means when someone offers to buy a thing. We did not elicit posterior beliefs before choice because asking whether a transfer signals anything about the item would itself have drawn attention to the hypothesis.\n\nThis is partly a measurement problem. The agreement changes the allocation of legal title and its description together. Variations that assigned titles under text held fixed, or changed the description while leaving title in one place, would answer different causal questions. The estimates here concern the complete offers. The small probability of service failure also leaves a contractual contingency. In its realized form the household received the same object and the same payment under a void in either arm. At intake the household might still have expected unequal failure rates because the form of the service looked more familiar under one agreement. We can compare the facts of delivery against that explanation. We cannot recover an expectation from the two failed batches.\n\nThe intended setting excludes an especially interesting group: households that retain a key because several members might know where it goes. An attendee had to have authority to decide, and any doubts about consent precluded enrollment. Some households may have interpreted authority as knowledge that another member would never need the thing. We did not interview absent household members to check that inference. A demand curve for a joint-ownership contract would include the cost of securing agreement and could give a much different picture [@hanse2021]. The experiment also removes search costs from the last decision by presenting one contract on the spot. Traveling to the office and identifying an eligible object happen before randomization. A collection service that advertised one contract to all households could change who participates before it changes how many agree at the counter.\n\nUnder agency, 18 of 64 owners at zero were prepared to dispose of a key for the service alone. A positive fee may be justified for those willing to pay it, but the selected intake sample does not provide a revenue forecast. A transfer service would acquire property with a planned scrap disposition from a subset of owners. Its intake mix could change if goods of observable value were admitted, as could provider discretion and the value of title itself. @sauer2022 reports that bulk collectors often diagnose such changes in composition only after operating for several months. The narrower finding here concerns the pricing of the last transaction for a single object.\n\n# Conclusion {#sec:conclusion}\n\nAn inventory enumerator who treats a discarded key as evidence that it was worthless to the owner has to explain why the inferred value depends on the disposal document. We observe the variation at the moment of choice, before the service has any opportunity to affect the object. The key's hypothetical use enters the decision to keep it. The cost of agreeing to be the owner on whose behalf it is destroyed enters the decision to give it up. A survey that asks only which objects a household expects to use again has no place to record the second quantity.\n\n\\appendix\n\n# Operative documents {#sec:documents}\n\nThe wording below gives the common English master. Staff displayed the local language. The office name and the payment amount were substituted by the allocation program. At confirmation the document time and the item number were appended to the agreement. Until confirmation the item stayed on the participant's side of the intake counter.\n\nTable: Clauses shown above the choice buttons. The remainder of the agreement was common to both conditions. {#tbl:clauses}\n\n| Condition | Authorization |\n|:--|:--|\n| Agency | You authorize the office to dispose of this item on your behalf. The item belongs to you until it is destroyed. |\n| Transfer | You transfer ownership of this item to the office for disposal. The item belongs to the office from the time you accept this offer. |\n\nThe common clause specified permanent destruction of the item during the scheduled visit and placement of its remnants in the provider's scrap stream. The provider could neither sell nor reuse the intact item. The participant could not ask to have it returned after confirmation. If the provider could not finish during the visit, the office returned the item and settled only the participation payment. Under transfer a void also reinstated the household's title. Staff provided a receipt stating completion or void. Neither the receipt nor the intake script offered a forecast of whether the key could ever be matched to a lock. A provider operations study by @dovran2023 supplied the wording for the common void clause, with the settlement amount adapted for the present offers.\n\n# Design variance and exact test {#sec:variance}\n\nIndex the ten offer arms by $a$, let $n_{sa}=16$ be the arm size, and let $c_{sa}$ be the coefficient on the arm mean in a given linear calculation, including the site weight. Within a site $N_s=160$. Write $S_{sa}^2$ for the finite-population variance of potential outcomes under arm $a$ in site $s$, and $S_{s,c}^2$ for the finite-population variance of the unit-level contrast $\\sum_a c_{sa}Y_i(a)$. Complete randomization gives\n\n$$\n\\operatorname{Var}\\!\\left(\\sum_{s,a}c_{sa}\\bar Y_{sa}\\right)\n=\\sum_{s,a}\\frac{c_{sa}^2 S_{sa}^2}{n_{sa}}\n-\\sum_s\\frac{S_{s,c}^2}{N_s}. \\label{eq:var}\n$$\n\nThe expression follows by taking the covariance of two assigned-arm means in a site, which is minus the finite-population covariance of their corresponding potential outcomes divided by $N_s$. Across sites the assignments are independent. We use the unbiased arm-level sample variance $n_{sa}\\bar Y_{sa}(1-\\bar Y_{sa})/(n_{sa}-1)$ in the first term and omit the second term. For @eq:diff every cell has coefficient of magnitude $1/20$ and the variance estimate is 0.0012454. For the pointwise differences in @tbl:prices the nonzero coefficients have magnitude $1/4$. In the confidence interval for the bounds in @sec:money, the linear coefficients follow by assigning weight €2 to each included endpoint in the relevant integral sum. The lower and upper endpoint standard errors are €0.329 and €0.318. Taking two one-sided limits at critical value 1.96 gives the reported outer interval with at least nominal 95% coverage asymptotically under the normal approximation. No covariance estimate between the endpoints is needed for that calculation.\n\nFor the exact test, condition on the site and price groups that were assigned. Under the sharp null every household in a group has a fixed binary choice. If $k_{sj}$ of the 32 households accepted, the number $X_{sj}$ assigned to transfer among those acceptances has the hypergeometric distribution with population size 32, $k_{sj}$ marked units, and 16 draws. The allocation rule makes these draws independent across groups conditional on the price assignments. The raw contrast under the null has the distribution of $\\sum_{sj}(2X_{sj}-k_{sj})$. We convolve the 20 integer-valued mass functions and sum the probabilities for absolute contrasts at least 49. The procedure accommodates each site's different base acceptance even under the sharp null. It does not treat a sharp hypothesis of zero effect for everyone as equivalent to an average null under heterogeneous effects.\n\n# Cell counts and analysis inputs {#sec:counts}\n\nTable: Acceptances by site and payment. Each cell gives agency then transfer, separated by a slash. Each numerator has denominator 16. Sites are in the order used in @fig:sites. {#tbl:cells}\n\n| Site | −€4 | −€2 | €0 | +€2 | +€4 |\n|:--|--:|--:|--:|--:|--:|\n| Lille | 1 / 3 | 3 / 5 | 5 / 9 | 7 / 10 | 11 / 13 |\n| Ghent | 2 / 2 | 3 / 5 | 4 / 8 | 8 / 11 | 10 / 13 |\n| Utrecht | 1 / 3 | 2 / 4 | 4 / 8 | 8 / 11 | 11 / 12 |\n| Aachen | 2 / 2 | 3 / 6 | 5 / 8 | 7 / 11 | 11 / 13 |\n\nWith equal cell sizes, @tbl:cells suffices to reproduce the acceptance curves, test, and interval calculations without household records. For @eq:clip the sum based on left endpoints is $2$ times the sum of the first four probabilities and gives the upper restricted mean. The corresponding sum over the last four endpoints gives the lower restricted mean. An interval estimate for the difference requires subtracting the upper transfer mean from the lower agency mean for the lower limit, and the lower transfer mean from the upper agency mean for the upper limit. Interpolating first is a separate calculation. It weights the two extreme payments by €1 apiece and the interior payments by €2, the trapezoid rule for the interpolated curve. All tabulated results use unrounded cell proportions from @tbl:cells. The household files contain booking details and free-text origins that remain with the intake offices. No part of the acceptance analysis requires those fields.\n","bibliography":"@article{coase1960,\n  author = {Coase, Ronald H.},\n  title = {The Problem of Social Cost},\n  journal = {Journal of Law and Economics},\n  volume = {3},\n  pages = {1--44},\n  year = {1960},\n  doi = {10.1086/466560}\n}\n@book{barzel1997,\n  author = {Barzel, Yoram},\n  title = {Economic Analysis of Property Rights},\n  edition = {2},\n  publisher = {Cambridge University Press},\n  year = {1997}\n}\n@article{akerlof1970,\n  author = {Akerlof, George A.},\n  title = {The Market for {Lemons}: Quality Uncertainty and the Market Mechanism},\n  journal = {The Quarterly Journal of Economics},\n  volume = {84},\n  number = {3},\n  pages = {488--500},\n  year = {1970},\n  doi = {10.2307/1879431}\n}\n@article{arrow1974,\n  author = {Arrow, Kenneth J. and Fisher, Anthony C.},\n  title = {Environmental Preservation, Uncertainty, and Irreversibility},\n  journal = {The Quarterly Journal of Economics},\n  volume = {88},\n  number = {2},\n  pages = {312--319},\n  year = {1974},\n  doi = {10.2307/1883074}\n}\n@book{dixit1994,\n  author = {Dixit, Avinash K. and Pindyck, Robert S.},\n  title = {Investment under Uncertainty},\n  publisher = {Princeton University Press},\n  year = {1994}\n}\n@article{thaler1980,\n  author = {Thaler, Richard},\n  title = {Toward a Positive Theory of Consumer Choice},\n  journal = {Journal of Economic Behavior and Organization},\n  volume = {1},\n  number = {1},\n  pages = {39--60},\n  year = {1980},\n  doi = {10.1016/0167-2681(80)90051-7}\n}\n@article{kahneman1990,\n  author = {Kahneman, Daniel and Knetsch, Jack L. and Thaler, Richard H.},\n  title = {Experimental Tests of the Endowment Effect and the {Coase} Theorem},\n  journal = {Journal of Political Economy},\n  volume = {98},\n  number = {6},\n  pages = {1325--1348},\n  year = {1990},\n  doi = {10.1086/261737}\n}\n@article{samuelson1988,\n  author = {Samuelson, William and Zeckhauser, Richard},\n  title = {Status Quo Bias in Decision Making},\n  journal = {Journal of Risk and Uncertainty},\n  volume = {1},\n  pages = {7--59},\n  year = {1988},\n  doi = {10.1007/BF00055564}\n}\n@article{plott2005,\n  author = {Plott, Charles R. and Zeiler, Kathryn},\n  title = {The Willingness to Pay--Willingness to Accept Gap, the Endowment Effect, Subject Misconceptions, and Experimental Procedures for Eliciting Valuations},\n  journal = {American Economic Review},\n  volume = {95},\n  number = {3},\n  pages = {530--545},\n  year = {2005},\n  doi = {10.1257/0002828054201387}\n}\n@article{list2003,\n  author = {List, John A.},\n  title = {Does Market Experience Eliminate Market Anomalies?},\n  journal = {The Quarterly Journal of Economics},\n  volume = {118},\n  number = {1},\n  pages = {41--71},\n  year = {2003},\n  doi = {10.1162/00335530360535144}\n}\n@article{hanemann1984,\n  author = {Hanemann, W. Michael},\n  title = {Welfare Evaluations in Contingent Valuation Experiments with Discrete Responses},\n  journal = {American Journal of Agricultural Economics},\n  volume = {66},\n  number = {3},\n  pages = {332--341},\n  year = {1984},\n  doi = {10.2307/1240800}\n}\n@book{imbens2015,\n  author = {Imbens, Guido W. and Rubin, Donald B.},\n  title = {Causal Inference for Statistics, Social, and Biomedical Sciences: An Introduction},\n  publisher = {Cambridge University Press},\n  year = {2015}\n}\n@book{manski2003,\n  author = {Manski, Charles F.},\n  title = {Partial Identification of Probability Distributions},\n  publisher = {Springer},\n  year = {2003}\n}\n@article{loret2018,\n  author = {Loret, Isabeau and M'Baye, Salif-Elias},\n  title = {Negative Prices and the Organization of Household Clearance},\n  journal = {European Journal of Household Accounts},\n  volume = {12},\n  number = {2},\n  pages = {85--113},\n  year = {2018}\n}\n@techreport{sauer2022,\n  author = {Sauer, Friederike K. and Tomic, Borjan E.},\n  title = {Costs and Composition at Municipal Small-Article Depots},\n  institution = {Westmark Institute for Local Public Finance},\n  number = {22-14},\n  year = {2022}\n}\n@article{sevrin2019,\n  author = {Sevrin, C\\'eline A. and Suljic, Mina L.},\n  title = {Sorting Rights and the Price of Mixed Household Lots},\n  journal = {Review of Secondary Markets},\n  volume = {7},\n  number = {1},\n  pages = {37--68},\n  year = {2019}\n}\n@article{neve2020,\n  author = {Neve, Kristine van de and Ofori, Samuel N.},\n  title = {Replacement Expenditure and Unclassified Stocks in Household Inventories},\n  journal = {European Journal of Household Accounts},\n  volume = {14},\n  number = {3},\n  pages = {173--205},\n  year = {2020}\n}\n@techreport{hanse2021,\n  author = {Hanse, Mette R. and Ionescu, Petra-Anca},\n  title = {Household Agreement and the Retirement of Shared Durables},\n  institution = {Ravnsholt Centre for Consumption Research},\n  number = {37},\n  year = {2021}\n}\n@techreport{dovran2023,\n  author = {Dovran, Mirela and Kappel, Antonin E.},\n  title = {Acceptance Records and Voided Dispositions in Scheduled Intake Services},\n  institution = {Institut des Services de Proximit\\'e},\n  number = {2023-06},\n  year = {2023}\n}","figures":[{"id":"site-contrasts","kind":"chart","data":"{\"type\":\"bar\",\"xlabel\":\"Intake site\",\"ylabel\":\"Transfer effect (percentage points)\",\"x\":[\"Lille\",\"Ghent\",\"Utrecht\",\"Aachen\"],\"series\":[{\"label\":\"Site estimate\",\"y\":[16.25,15,15,15],\"err\":[13.9,13.86,13.63,13.95]}],\"hlines\":[{\"y\":15.3125,\"style\":\"dashed\"}],\"ylim\":[0,34],\"legend\":\"none\",\"size\":[5.2,2.9],\"style\":\"pgfplots\"}"}],"acknowledgments":"We thank Céline A. Sevrin for reviewing the intake documents and the four office registrars for reconciling the service ledgers. The Ferein Trust for Research on Consumption supported the fieldwork.","options":{"citation_style":"author-year"}}