1# RECORD 4a6c4412ae803b09046980401041b08141e48e512# TITLE: 'No Drugs in My Back Yard:' the Ambivalent Reception of Cannabis Retailers3# YEAR: 2019.0 VENUE: Social Science Research Network DOI: 10.2139/ssrn.34618204# AUTHORS: L. M. Bruijn; R. Ribas5# PAGES: 20 CANDIDATE TABLE PAGES (1-based): [9, 12, 13, 14, 15, 16]6# ABSTRACT: Can individuals' aversion to drug markets curb the benefits of decriminalization? We investigate the effect of two policies on housing demand in the Netherlands: the distance-to-school criterion, which closed some cannabis shops in a few cities; and the zero-tolerance policy, which banned shops within municipal jurisdictions. While a small increase in the distance to retailers raised house prices by 1-5%, a substantial increase reduced them by 1-6%. Both policies reduced property crime, but the zero-tolerance was also related to fewer jobs. Our findings reveal that cities benefit from having cannabis shops, but households' aversion to related nuisances depreciates surrounding areas.789===== PAGE 1 (score 4, numeric-paren 0, table-words 0, var-words 4) =====1011University of Groningen12“No drugs in my back yard”13Bruijn, Larissa Michelle; Ribas , Rafael P.14Published in:15Journal of Economic Behavior & Organization16DOI:1710.1016/j.jebo.2022.05.00518IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from19it. Please check the document version below.20Document Version21Publisher's PDF, also known as Version of record22Publication date:23202224Link to publication in University of Groningen/UMCG research database25Citation for published version (APA):26Bruijn, L. M., & Ribas , R. P. (2022). “No drugs in my back yard”: The ambivalent reception of cannabis27retailers. Journal of Economic Behavior & Organization, 199, 103-121.28https://doi.org/10.1016/j.jebo.2022.05.00529Copyright30Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the31author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).32The publication may also be distributed here under the terms of Article 25fa of the Dutch Copyright Act, indicated by the “Taverne” license.33More information can be found on the University of Groningen website: https://www.rug.nl/library/open-access/self-archiving-pure/taverne-34amendment.35Take-down policy36If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately37and investigate your claim.38Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the39number of authors shown on this cover page is limited to 10 maximum.40Download date: 21-10-2026414243===== PAGE 9 (score 45, numeric-paren 24, table-words 5, var-words 7) =====4445L.M. Bruijn and R.P. Ribas 46Journal of Economic Behavior and Organization 199 (2022) 103–121 47Table 1 48Cross-Sectional Relationship between House Prices and Distance to Retailer. 49Period 501999–2001 512012–2016 52(1) 53(2) 54(3) 55(4) 56(5) 57(6) 58Distance in 1999 (miles) < 0 . 2 590.044 ∗∗∗60−0.036 610.056 ∗∗∗62−0.030 63(0.014) 64(0.035) 65(0.016) 66(0.035) 670.2–0.4 680.024 ∗∗69−0.039 700.048 ∗∗∗71−0.032 72(0.011) 73(0.030) 74(0.011) 75(0.030) 760.4–1.0 770.015 ∗78−0.039 ∗∗790.032 ∗∗∗80−0.035 ∗81(0.008) 82(0.019) 83(0.009) 84(0.020) 85Distance in 2016 (miles) 86< 0 . 2 870.057 ∗∗∗880.090 ∗∗890.067 ∗∗∗900.094 ∗∗91(0.017) 92(0.041) 93(0.019) 94(0.042) 950.2–0.4 960.035 ∗∗∗970.071 ∗∗980.060 ∗∗∗990.088 ∗∗100(0.013) 101(0.035) 102(0.015) 103(0.038) 1040.4–1.0 1050.027 ∗∗∗1060.063 ∗∗∗1070.046 ∗∗∗1080.076 ∗∗∗109(0.009) 110(0.021) 111(0.010) 112(0.023) 113House characteristics 114Yes 115Yes 116Yes 117Yes 118Yes 119Yes 120Year dummies 121Yes 122Yes 123Yes 124Yes 125Yes 126Yes 127Town fixed effects 128Yes 129Yes 130Yes 131Yes 132Yes 133Yes 134Number of observations 135171,804 136171,804 137171,804 138300,967 139300,967 140300,967 141Number of towns 142201 143201 144201 145201 146201 147201 148This table presents the linear regression of log price per square meter of traded houses on their shortest distance to a cannabis retailer. This distance is 149represented by three dummies: less than 0.2 miles, between 0.2 and 0.4 miles, and between 0.4 and one mile; the reference is the group of houses farther 150than one mile. Columns (1) to (3) show estimates for transaction prices from 1999 to 2001 as a function of distance in 1999, distance in 2016, and distances 151in both 1999 and 2016, respectively. Columns (4) to (6) show the same regressions but from 2012 to 2016. All regressions include house characteristics, as 152in Table A1, year dummies, and town fixed effects. The sam ple includes properties located up to 3.1 miles from the nearest retailer in towns that have ever 153had a retailer within one mile. Clustered robust standard errors are in parentheses. ∗∗∗, ∗∗, ∗represent statistical significance at the 1%, 5%, and 10% levels, 154respectively. 155We also estimate the relationship between demographic characteristics and the distance to a retailer. Table A3 of the 156Online Appendix displays the estimates. In summary, areas near coffeeshops concentrate more men, people between 15 and 15739 years old, single-person households, and Western immigrants. In contrast, these areas present relatively fewer children 158and adults over 40. The closures were more often in areas with a lower share of adults under 40, higher shares of adults 159over 40 and non-Western immigrants, and more elderly people in the vicinity. Over time, those areas have experienced an 160outflow of adults under 40, compensated by an inflow of adults over 40 and Western immigrants. 161These initial findings stress the importance of controlling for local characteristics, as well as using a comparison group 162that is equally distant to a retailer, when estimating the effect of closures. Presumably, the postcode fixed effects should 163control for the selection of areas with closed retailers, and the comparison group should control for the overall appreciation 164of areas with a coffeeshop. If specific variations in property values triggered the closure policies, we should observe this 165variation before treatment (see Fig. 3 below). 1665. Empirical strategy 167This section describes our empirical models. First, it presents the reduced-form difference-in-differences (DiD), which 168compares houses whose nearest retailer was closed with houses that were equally distant from a retailer. For this compari- 169son, we consider only closures forced by the distance-to-school criterion (DTSC) and the zero-tolerance policy (ZTP). Homes 170whose nearest shop was closed for another reason are excluded from the analysis. 171The closures made by those policies result in two instrumental variables (IVs): nearest shop closed by the DTSC and 172nearest shop closed by the ZTP. While the former instrument is used to identify the effect of a small but significance change 173in the distance to retailers, the latter is used to identify the effect of a much larger increase in distance. Accordingly, we 174apply an IV strategy, described in the second part of this section, to estimate the nonlinear effect of distance to cannabis 175outlets on house prices. Other empirical models used in this study are described in the Online Appendix. 1765.1. Reduced form 177Closing a coffeeshop should, if anything, increase the shortest distance from homes to retailers. Thus, we first estimate 178the effect that closures have on house prices. Let y ilt be the log price per square meter of house i in location l and year t, c l 179be the indicator whether the nearest retailer is closed, and q l be the year of closure. The average effect is identified by the 180following DiD estimator: 181τ = E(y ilt | c l = 1 , t ≥q l ) −E(y ilt | c l = 0 , t ≥q l ) 182−[ E(y ilt | c l = 1 , t < q l ) −E(y ilt | c l = 0 , t < q l )] 183(1) 184110 185186187===== PAGE 10 (score 5, numeric-paren 0, table-words 1, var-words 3) =====188189L.M. Bruijn and R.P. Ribas 190Journal of Economic Behavior and Organization 199 (2022) 103–121 191Fig. 3. Estimated Differences in House Prices around the Year of Closure. This figure presents the lead and lagged difference-in-differences (DiD) on the 192log price per square meter of traded houses. Panel (a) shows the changes before and after cannabis retailers are closed by the distance-to-school criterion 193(DTSC). Panel (b) shows the changes before and after introduction of the zero-tolerance policy. Each graph uses a different sam ple, defined by the initial 194distance to a retailer. All regressions include house characteristics, as in Table A1, year dummies, year dummies times a dummy for municipalities that 195implemented the DTSC, and six-digit postcode fixed effects. The sample excludes houses in which the nearest retailer was closed for another reason. 196Robust standard errors are clustered by postcode. Vertical lines around the point estimates represent 90% and 95% confidence intervals, respectively. 197where c l is one if the nearest retailer is closed and zero otherwise. Location refers to the centroid of a six-digit postcode. 198The DiD parameter, τ, is estimated using a fixed-effect model and varies with the initial distance to a cannabis shop: 199y ilt = 200⎧ 201⎪ 202⎪ 203⎪ 204⎨ 205⎪ 206⎪ 207⎪ 208⎩ 209τ 0 c l · I(t ≥q l ) + β0 x i + θ 0 gdp lt + μ0 210t + μl + ε ilt 211if d l0 ∈ 2120 , D 0 213τ 1 c l · I(t ≥q l ) + β1 x i + θ 1 gdp lt + μ1 214t + μl + ε ilt 215if d l0 ∈ 216D 0 , D 1 217. . . 218τ n c l · I(t ≥q l ) + βn x i + θ n gdp lt + μn 219t + μl + ε ilt 220if d l0 ∈ 221D n −1 , D n 222, 223(2) 224where x i is a vector of house characteristics, gdp lt is the regional log GDP, { μ j 225t } n 226j=0 are year-specific effects, μl is the postcode 227fixed effect, ε ilt is the deal-specific error term, d l0 is the initial distance from location l to the nearest shop, and { D j } n 228j=0 is 229a set of distance thresholds — say 0.2, 0.3, 0.4, 0.6, and 0.8 miles. Standard errors are clustered by postcode to account for 230serial error correlation ( Bertrand et al., 2004 ). 231Since the regulation of sales and the demand for cannabis have evolved, the (dis)amenity created by retailers has also 232varied. Even if the distance of a home to retailers remains the same, its price may respond to changes in nuisances. To 233control for time-varying unobservables, we let the year-specific effects, { μ j 234t } n 235j=0 , vary across location ranges so that they 236capture distinct improvements happening near and far away from cannabis outlets. This type of DiD is different from those 237applied by, for instance, Linden and Rockoff (2008) and Conklin et al. (2017) , who instead compare near and distant areas 238over time. By breaking the effect into several location ranges, we can also verify the magnitude of general equilibrium effects 239on further areas. 240111 241242243===== PAGE 12 (score 37, numeric-paren 16, table-words 6, var-words 6) =====244245L.M. Bruijn and R.P. Ribas 246Journal of Economic Behavior and Organization 199 (2022) 103–121 247Table 2 248Estimated Effect of Closures on Distance to Retailer and House Price. 249Initial distance to retailer (miles) 2500–0.2 2510.2–0.3 2520.3–0.4 2530.4–0.6 254Distance to retailer (miles) 255Distance-to-school 2560.147 ∗∗∗2570.118 ∗∗∗2580.127 ∗∗∗2590.162 ∗∗∗260(0.010) 261(0.011) 262(0.020) 263(0.026) 264Zero-tolerance 2653.562 ∗∗∗2663.070 ∗∗∗2673.172 ∗∗∗2683.416 ∗∗∗269(0.161) 270(0.221) 271(0.173) 272(0.117) 273Log price per m 2 274Distance-to-school 2750.015 2760.041 ∗∗2770.053 ∗∗278−0.001 279(0.017) 280(0.020) 281(0.021) 282(0.020) 283Zero-tolerance 284−0.024 285−0.059 ∗∗∗286−0.009 287−0.007 288(0.015) 289(0.021) 290(0.020) 291(0.012) 292House characteristics 293Yes 294Yes 295Yes 296Yes 297Regional GDP 298Yes 299Yes 300Yes 301Yes 302Year dummies 303Yes 304Yes 305Yes 306Yes 307Year dummies × DTSC 308Yes 309Yes 310Yes 311Postcode fixed effects 312Yes 313Yes 314Yes 315Yes 316Number of observations 31773,937 31850,368 31945,659 32094,497 321Number of postcodes 3228561 3235445 3245044 32510,504 326with distance-to-school 327479 328210 32998 33083 331with zero-tolerance 332284 333210 334228 335484 336comparison 3377798 3385025 3394718 3409937 341This table presents the difference-in-differences (DiD) estimates, as in Eq. (2) , for two outcomes: shortest distance to a cannabis retailer (top panel) and 342log price per square meter of traded houses (bottom panel). Each regression contains two DiDs: one for closures forced by the distance-to-school criterion 343(DTSC), and another for closures forced by the zero-tolerance policy. Each column shows a regression using a different sample, defined by the initial 344distance to a retailer. All regressions include house characteristics, as in Table A1, year dummies, year dummies times a dummy for municipalities that 345implemented the DTSC, and six-digit postcode fixed effects. The sample excludes houses in which the nearest retailer was closed for another reason. 346Clustered robust standard errors are in parentheses. ∗∗∗, ∗∗, ∗represent statistical significance at the 1%, 5%, and 10% levels, respectively. 347Conklin et al. (2017) , who show that prices increase by 8% after medical dispensers are converted into retailers, and with 348Cheng et al. (2018) , who show a rise of 6% in municipalities in Colorado that legalized marijuana retail. 349Nevertheless, if there are many retailers in the city, closing some of them reshuffles the housing demand, raising prices in 350adjacent neighborhoods and reducing prices further away. Although the city benefits from having cannabis retailers, house- 351holds seem to pay to be just far enough from them. The latter is in line with Adda et al. ’s (2014) evidence that the de- 352criminalization of cannabis possession decreases house prices in hot spots and increases them in areas where it is still a 353crime. 354As regards the consistency of our findings, one may argue that they result from differences in the previous trends 355between places where shops were closed and where they were not — i.e., a violation of the parallel trend assumption 356( Abadie, 2005 ). Fig. 3 shows the difference in log prices between treated and comparison groups for each year before and 357after the closure of the nearest shop. Before, the difference fluctuates around zero and does not exhibit a distinct trend for 358houses affected by the DTSC and by the ZTP. 359After the nearest coffeeshop closes due to the DTSC, prices immediately increase between 0.3 and 0.4 miles away and 360persist at least until the third year. The positive effect on homes between 0.2 and 0.3 miles appears as significant only four 361years later. 15 Therefore, it takes a few years for the housing demand to adjust to the new policy fully. 362Similarly, the effect of a ZTP seems gradual. For homes within 0.3 miles, the effect appears at least a year after the last 363coffeeshop was closed in the municipality. For homes between 0.4 and 1 miles away, we observe a short-term effect that 364does not persist after three years. Overall, the ZTP shifts housing demand in different areas and with different timings, but 365the net impact on the local economy ends up negative after three years. 3666.2. The estimated distance function 3676.2.1. House prices 368The findings in the previous part suggest that the relationship between house prices and distance to cannabis shops is 369non-monotonic. That is, a small increase in distance is followed by an increase in prices, while a large increase in distance 370depreciates prices. Table 3 presents the estimated distance function, as in Eq. (3) , using two estimators. The first one takes 371into account changes in distance provoked by any closure. Then we restrict the sample to homes whose distance to the 372nearest and second-nearest shop has not changed, and those whose distance was affected by either policy. The closures 37315 The higher significance of the effect for this range in Table 2 is due to the aggregation of values in the post-treatment period, which produces narrower 374confidence intervals. 375113 376377378===== PAGE 13 (score 46, numeric-paren 16, table-words 11, var-words 5) =====379380L.M. Bruijn and R.P. Ribas 381Journal of Economic Behavior and Organization 199 (2022) 103–121 382Table 3 383Estimated Effect of Distance to Retailer on House Price. 384Final 385Initial distance to retailer (miles) 386distance (miles) 387< 0 . 3 388< 0 . 4 389< 0 . 5 390< 0 . 6 391Least square 392< 0 . 3 393−0.016 ∗∗∗394−0.013 ∗∗∗395−0.015 ∗∗∗396−0.016 ∗∗∗397(0.004) 398(0.004) 399(0.004) 400(0.004) 401> 2 . 0 402−0.056 ∗∗∗403−0.034 ∗∗∗404−0.028 ∗∗∗405−0.024 ∗∗∗406(0.009) 407(0.008) 408(0.006) 409(0.006) 410House characteristics 411Yes 412Yes 413Yes 414Yes 415Regional GDP 416Yes 417Yes 418Yes 419Yes 420Year dummies 421Yes 422Yes 423Yes 424Yes 425Year dummies × DTSC 426Yes 427Yes 428Yes 429Yes 430Postcode fixed effects 431Yes 432Yes 433Yes 434Yes 435Number of observations 436199,172 437274,769 438350,269 439422,630 440Number of postcodes 44122,087 44230,268 44338,518 44446,630 445Instrumental variable 446< 0 . 3 447−0.019 448−0.046 ∗∗449−0.064 ∗∗∗450−0.067 ∗∗∗451(0.021) 452(0.021) 453(0.022) 454(0.023) 455> 2 . 0 456−0.068 ∗∗457−0.060 ∗∗∗458−0.058 ∗∗∗459−0.047 ∗∗∗460(0.029) 461(0.021) 462(0.016) 463(0.013) 464House characteristics 465Yes 466Yes 467Yes 468Yes 469Regional GDP 470Yes 471Yes 472Yes 473Yes 474Year dummies 475Yes 476Yes 477Yes 478Yes 479Year dummies × DTSC 480Yes 481Yes 482Yes 483Yes 484Postcode fixed effects 485Yes 486Yes 487Yes 488Yes 489LM stat. for under-identification 490155.1 491148.8 492145.5 493143.6 494Wald stat. for weak instruments 49598.9 49692.9 49787.7 49886.3 499Number of observations 500122,605 501167,757 502214,215 503261,261 504Number of postcodes 50512,306 50616,843 50721,530 50826,354 509This table presents the relationship between changes in the distance to the nearest cannabis retailer and in the log of house price per square meter, as in 510Eq. (3) . The distance is represented by two dummies: less than 0.3 miles, and more than two miles; the reference is the group of houses between 0.3 and 511two miles. All regressions include house characteristics, as in Table A1, year dummies, year dummies times a dummy for municipalities that implemented 512the distance-to-school criterion (DTSC), and postcode fixed effects. The columns show the results from different samples, based on their initial distance 513to a retailer. In the top panel, regressions exploit any variation in the distance. In the bottom panel, variations in the distance are instrumented by the 514closures forced by the DTSC and the zero-tolerance policy. The sample in the bottom panel excludes houses in which the nearest retailer was closed for 515another reason. Clustered robust standard errors are in parentheses. ∗∗∗, ∗∗, ∗represent statistical significance at the 1%, 5%, and 10% levels, respectively. 516made by the two policies are used as IVs for the distance function. In both models, we control for house characteristics, 517postcode fixed effects, regional GDP, and year dummies. 518Since both house prices and the operation of coffeeshops can jointly respond to other local events, the variation in 519distance can be endogenous in the first estimated model (in the top panel of Table 3 ). With the IVs, though, changes in 520distance are triggered by exogenous events, which are not preceded by variations in house prices (see Fig. 3 ). Indeed, the 521least-square estimates point to the same conclusion as the IV estimates, but the magnitudes are at most half as large. Fig. A2 522of the Online Appendix confirms that closures brought about for procedural reasons also increase the price of properties 523farther than 0.2 miles. However, it also indicates that prices farther than 0.4 miles were increasing before these closures. 524The IV estimates in Table 3 reveal that houses are 2–7% less valuable if within 0.3 miles of retailers, but they are also 5–5257% cheaper if located farther than two miles. In other words, house prices peak at a distance at which cannabis shops are far 526enough to avoid any disturbance, but they decrease moving farther. This result is robust to different distance categorizations; 527see Table A6 of the Appendix. If we expand the definition of “near” to within 0.4 miles, the magnitudes of IV estimates 528become much larger. However, the IVs get weaker. Thus, we choose to present more conservative estimates as our main 529results. Section 7 examines some reasons for the non-monotonicity in housing prices. 530We also estimate the distance function with a restricted cubic spline. Fig. A3 of the Online Appendix suggests that prices 531are 15–25% higher about two miles from the retailers, and decline moving farther away. This pattern is similar to the one 532Giambona and Ribas (2018) report for prostitution in the Netherlands. Their evidence is that households pay an increasing 533premium to live as far as two kilometers from areas with prostitution. 5346.2.2. Demographic composition 535A critical concern is that the non-monotonic pattern resulted from the sorting of distinct demographic groups after the 536closures. That is, the households driving values down around retailers were different from those affecting values far away. To 537verify this issue, we estimate the share of demographic groups using the same IV model as in Table 3 . For most groups, the 538estimates are not significant. In Table 4 , we present those that are. Table A7 of the Online Appendix presents the remaining 539results. 540114 541542543===== PAGE 14 (score 62, numeric-paren 32, table-words 7, var-words 10) =====544545L.M. Bruijn and R.P. Ribas 546Journal of Economic Behavior and Organization 199 (2022) 103–121 547Table 4 548Estimated Effect of Distance to Retailer on Demographic Composition. 549Final 550Initial distance to retailer (miles) 551distance (miles) 552< 0 . 3 553< 0 . 4 554< 0 . 5 555< 0 . 6 556Log share of 25–39 years old 557< 0 . 4 558−0.515 ∗559−0.466 ∗∗560−0.527 ∗561−0.545 ∗∗562(0.273) 563(0.236) 564(0.275) 565(0.277) 566> 2 . 0 567−0.462 ∗∗568−0.347 ∗∗569−0.394 ∗570−0.361 ∗571(0.222) 572(0.169) 573(0.211) 574(0.201) 575Log share of 60 + years old 576< 0 . 4 5770.514 ∗5780.465 ∗5790.567 ∗5800.590 ∗∗581(0.306) 582(0.251) 583(0.290) 584(0.293) 585> 2 . 0 5860.502 ∗5870.395 ∗∗5880.454 ∗∗5890.411 ∗590(0.260) 591(0.181) 592(0.223) 593(0.214) 594Log share of single-person households 595< 0 . 4 5960.204 ∗∗5970.209 ∗∗5980.288 ∗∗5990.327 ∗∗600(0.095) 601(0.101) 602(0.140) 603(0.155) 604> 2 . 0 6050.165 ∗∗6060.159 ∗∗6070.230 ∗∗6080.236 ∗∗609(0.077) 610(0.070) 611(0.105) 612(0.109) 613Log share of couples with children 614< 0 . 4 615−0.323 ∗616−0.286 ∗617−0.331 ∗∗618−0.341 ∗∗619(0.176) 620(0.150) 621(0.167) 622(0.166) 623> 2 . 0 624−0.288 ∗∗625−0.248 ∗∗626−0.270 ∗∗627−0.250 ∗∗628(0.143) 629(0.105) 630(0.124) 631(0.118) 632Year dummies 633Yes 634Yes 635Yes 636Yes 637Neighborhood fixed effects 638Yes 639Yes 640Yes 641Yes 642Number of observations 6432182 6443226 6454285 6465264 647Number of neighborhood 648151 649211 650280 651338 652This table presents the relationship between changes in the distance to the nearest cannabis retailer and in the log share of certain demographic groups 653in the neighborhood. Results for other groups are presented in Table A7 of the Online Appendix. A neighborhood is defined by its four-digit postcode. 654The distance is represented by two dummies: less than 0.4 miles, and more than two miles; the reference is the group of houses between 0.4 and two 655miles. All regressions include year dummies and neighborhood fixed effects, and the dummies for distance are instrumented by the closures caused by the 656distance-to-school criterion and the zero-tolerance policy. The columns show the results from different samples, based on their initial distance to a retailer. 657The sample excludes neighborhoods in which the nearest retailer was closed for another reason. Clustered robust standard errors are in parentheses. ∗∗∗, 658∗∗, ∗represent statistical significance at the 1%, 5%, and 10% levels, respectively. 659As regards the demographics, adults under 40 and couples with children are the least willing to live either near or far 660away from shops. In contrast, elderly and single people are the most willing to live in these areas. Despite these differences, 661these results indicate that the same demographic groups explain the results in Table 3 . 6627. Additional results 663In this section, we present additional results to help us understand the ambivalent effect of closure policies on house 664prices. First, we show that these policies affect the whole distribution of prices, yet the effect is higher in the lower tail. 665Then we verify whether our main results are explained by changes in housing supply or other amenities, such as bars and 666schools. Finally, we investigate what also happened in terms of employment and crime rates after closures. 6677.1. Estimated effects on the distribution of prices 668From the previous results, one can make inferences on the average capitalization effect of closing retailers ( Kuminoff and 669Pope, 2014 ). This average should lie between former residents’ and future residents’ willingness to pay for coffeeshops (after 670closures), as long as the comparison group is not affected by those policies ( Banzhaf, 2018 ). To shed some light on the 671distribution of tastes, we also estimate the effect on conditional quantiles of house prices, which reveals the house’s risk of 672under- and overvaluation. Details are in Section 11 of the Online Appendix. 673Fig. 4 presents the estimated quantile effects of closures. The left-hand graph shows the effect caused by the DTSC, 674for homes between 0.2 and 0.4 miles. The right-hand graph shows the effect of a ZTP, for homes within 0.3 miles. These 675distance ranges are defined based on the results from Table 2 . In both cases, the effect is higher in the lower tail. 676The DTSC is found to shift the entire distribution of prices, raising all quantiles. Overall, those households agree that 677being 0.13 miles more distant is beneficial. Besides, the higher effect at the lower tail means that a house is less likely to 678be underpriced. That is, the DTSC made it easier to find a buyer. The effect of the ZTP has the opposite pattern. Households 679living within 0.3 miles agree that the loss of coffeeshops is detrimental. Moreover, those houses are more likely to be 680underpriced — i.e., it became harder to find a buyer. 681To complement this analysis, we also estimate the effects on the number of days a house stays on the market. Table A5 682of the Online Appendix displays these estimates. Indeed, they confirm that after the DTSC, houses are sold more quickly. 683The ZTP presents similar effects, but they are not found to be significant. 684115 685686687===== PAGE 15 (score 37, numeric-paren 12, table-words 8, var-words 7) =====688689L.M. Bruijn and R.P. Ribas 690Journal of Economic Behavior and Organization 199 (2022) 103–121 691Fig. 4. Quantile Effects of Closings on House Prices. This figure presents the quantile regression coefficients on the residual of a hedonic model, which 692includes house characteristics, as in Table A1, regional GDP, year dummies, year dummies times a dummy for municipalities that implemented the distance- 693to-school criterion (DTSC), and six-digit postcode fixed effects. See equation (K.1) in the Online Appendix. In the left-hand graph, the sample includes 694properties located between 0.2 and 0.4 miles from the nearest retailer. In the right-hand graph, the sample includes properties located within 0.3 miles 695of the nearest retailer. Both samples exclude houses in which the nearest retailer was closed for another reason. The graphs come from separate quantile 696regressions, which have a simple difference-in-differences specification. The grey area represents the 95% confidence interval of the coefficients. 697Table 5 698Estimated Effect of Closing Policies on Housing Stock. 699Initial distance to retailer (miles) 700< 0 . 3 701< 0 . 4 702< 0 . 6 703< 0 . 8 704< 1 . 0 705< 3 . 0 706Distance-to-school 707−0.011 708−0.016 709−0.012 710−0.012 711−0.011 712−0.001 713(0.014) 714(0.014) 715(0.012) 716(0.012) 717(0.011) 718(0.009) 719Number of observations 7201547 7211781 7222332 7232813 7243313 7255438 726Number of cells 727119 728137 729181 730218 731258 732424 733Zero-tolerance 734−0.007 7350.001 7360.001 737−0.001 7380.000 739−0.006 740(0.011) 741(0.011) 742(0.010) 743(0.008) 744(0.008) 745(0.006) 746Number of observations 74715,525 74820,065 74931,459 75043,711 75156,720 752211,362 753Number of cells 7541196 7551546 7562426 7573373 7584377 75916,383 760Year dummies 761Yes 762Yes 763Yes 764Yes 765Yes 766Yes 767Cell fixed effects 768Yes 769Yes 770Yes 771Yes 772Yes 773Yes 774This table presents the difference-in-differences (DiD) estimates, as in Eq. (2) , for the log number of homes in a 500m-by-500m cell. Two separate DiDs 775are run using Poisson regressions: one for closures forced by the distance-to-school criterion, and another for closures forced by the zero-tolerance policy. 776Each column uses a different sample, defined by the initial distance of the cell to a retailer. All regressions include year dummies and cell fixed effects. The 777sample excludes cells in which the nearest retailer was closed for another reason. The top panel includes cells only in Rotterdam and Amersfoort. Clustered 778robust standard errors are in parentheses. ∗∗∗, ∗∗, ∗represent statistical significance at the 1%, 5%, and 10% levels, respectively. 7797.2. Housing supply and other amenities 780Other than the effect of the closure policies on housing demand, our main findings could also be explained by changes in 781housing supply and other amenities. Regarding the housing supply, Table 5 shows that the number of houses near cannabis 782shops is not significantly affected. Given the inelastic supply, our results on house prices should be interpreted as shifts in 783housing demand. 784As regards the DTSC, the concern is that the relationship between house prices and distance to schools has changed over 785time. Table A8 of the Online Appendix shows the estimated effect controlling for this time-varying relationship. Indeed, the 786demand for houses closer to primary schools has increased in Rotterdam and Amersfoort. Nevertheless, the estimated effect 787of the closure policy remains nearly the same after controlling for this process. 788Another concern is whether our results were explained by other businesses that tend to cluster around coffeeshops, such 789as bars and nightclubs. Table A9 of the Appendix confirms that the ZTP slightly increased the distance of homes to these 790establishments. This evidence suggests that bars seem to cluster around cannabis retailers. Notwithstanding, the estimated 791effects on house prices do not change after we control for the time-varying distance to bars. 792116 793794795===== PAGE 16 (score 31, numeric-paren 10, table-words 4, var-words 11) =====796797L.M. Bruijn and R.P. Ribas 798Journal of Economic Behavior and Organization 199 (2022) 103–121 799Table 6 800Estimated Effect of Closures on Employment and Crime. 801jobs/ 802soft drug 803hard drug 804property 805inhabitant 806unemployment 807offenses 808offenses 809violence 810crimes 811Distance-to-school 812−0.659 8130.017 8140.048 815−0.238 ∗∗∗816(0.459) 817(0.208) 818(0.073) 819(0.059) 820Zero-tolerance 821−0.039 ∗∗∗8220.001 823−0.054 8240.238 ∗∗∗825−0.097 826−0.125 ∗∗827(0.013) 828(0.001) 829(0.080) 830(0.069) 831(0.140) 832(0.063) 833Regional GDP 834Yes 835Yes 836Yes 837Yes 838Yes 839Yes 840Municipal jobs and unemployment 841No 842No 843Yes 844Yes 845Yes 846Yes 847Year dummies × province 848Yes 849Yes 850Yes 851Yes 852Yes 853Yes 854Year dummies × Rotterdam 855No 856No 857Yes 858Yes 859Yes 860Yes 861Municipality fixed effects 862Yes 863Yes 864No 865No 866No 867No 868Neighborhood fixed effects 869No 870No 871Yes 872Yes 873Yes 874Yes 875No. of observations 876730 877876 8781090 8791150 8801150 8811150 882No. of municipalities/neighborhoods 88373 88473 885109 886115 887115 888115 889with distance-to-school 8900 8910 89232 89332 89432 89532 896with zero-tolerance 8973 8983 8994 9004 9014 9024 903comparison 90469 90569 90680 90780 90880 90980 910This table presents the difference-in-differences (DiD) estimates for labor outcomes and crime. Each regression contains two DiDs: one for closures forced 911by the distance-to-school criterion, and another for closures forced by the zero-tolerance policy. The first two columns show linear regressions using a 912sample of municipalities. The last four columns show Poisson regressions using a sample that combines municipalities and neighborhoods of Rotterdam. 913The samples exclude municipalities that have never had cannabis retailers or in which retailers were closed for another reason. Clustered robust standard 914errors are in parentheses. ∗∗∗, ∗∗, ∗represent statistical significance at the 1%, 5%, and 10% levels, respectively. 9157.3. Change in employment and crimes 916To understand the changes in housing demand, we finally look at other objective outcomes related to employment and 917nuisances. Table 6 presents the average DiD estimates for these outcomes. Fig. 5 shows the estimates per year before and 918after the closures. We use this figure to verify the timing of the effects and whether previous trends explain our results. 919The effects on employment outcomes are at the municipal level, so we only report the effects of the ZTP. The effect 920on crime rates, however, is estimated at the neighborhood level in Rotterdam. Thus, at least for Rotterdam, we compare 921areas where the DTSC closed shops with areas that still have coffeeshops. For other municipalities, crime rates are at the 922municipal level. 9237.3.1. Estimated effects on employment 924In terms of labor market outcomes, the ZTP had no effect on the unemployment rate. This null evidence is consistent 925with other studies that investigate the relationship between access to cannabis and labor outcomes (e.g., van Ours, 2006; 926Sabia and Nguyen, 2018 ). 16 927Although the employment of municipal residents does not change, the number of workers employed in the city gradually 928declines after the ban — see Fig. 5 . Banning cannabis retailers from the jurisdiction reduces by 4 percentage points (or 9%) 929the number of jobs per inhabitant. One explanation for this economic impact is that tight quantity constraints on any 930business should make the economy less efficient ( Buchanan and Tullock, 1975; Gruber, 2011 ). However, we cannot verify 931whether the economic loss is either the cause or the consequence (or both) of a depreciated housing market. 9327.3.2. Estimated effects on drug-related crimes 933Regarding drug consumption, we find that both policies reduced the number of crimes related to soft drugs. This effect, 934though, is not persistent and becomes insignificant four years later. On the other hand, the ZTP has another side effect. After 935closing the last cannabis outlet, crimes related to hard drugs soared by 24%. This effect is also increasing over time, reaching 936over 50% four years after introduction of the policy. The result suggests that legal access to soft drugs is not a gateway, but 937a substitute for hard drugs ( Model, 1993; van Ours, 2003; Chu, 2015 ). 17 Once again, we cannot tell whether the depreciated 938housing market and the declining number of jobs either cause or result from the rise in illicit drug offenses. 9397.3.3. Estimated effects on violence and property crimes 940Neither policy significantly affects violence, but both are found to reduce the number of thefts by at least 12%. Unlike the 941evidence from Chang and Jacobson (2017) , Hunt et al. (2018) , and Brinkman and Mok-Lamme (2019) , this result contradicts 942the “eyes on the street” role of cannabis retailers. Both studies find that marijuana dispensers reduced property crimes in 94316 DeSimone (2002) is one of a few who shows that more access leads to a lower employment rate. See also van Ours and Williams (2015) for a 944comprehensive review. 94517 DiNardo and Lemieux (2001) and Conlin et al. (2005) present a similar pattern regarding alcohol and illicit drugs, and Bradford and Brad- 946ford (2018) present evidence for the substitution between cannabis and prescription drugs. In contrast, Melberg et al. (2010) and Kelly and Ra- 947sul (2014) present evidence for the gateway effect. 948117 949950951===== PAGE 17 (score 4, numeric-paren 0, table-words 0, var-words 4) =====952953L.M. Bruijn and R.P. Ribas 954Journal of Economic Behavior and Organization 199 (2022) 103–121 955Fig. 5. Estimated Differences in Employment and Crimes around the Year of Closure. This figure presents the lead and lagged difference-in-differences 956(DiD) on labor outcomes and crime rates. Panel (a) shows the changes before and after cannabis retailers are closed by the distance-to-school criterion 957(DTSC). Panel (b) shows the changes before and after the zero-tolerance policy. The labor outcomes (the first two graphs in each panel) are estimated by 958a linear regression using a sample of municipalities. Crime rates are estimated by a Poisson regression using a sample that combines municipalities and 959neighborhoods of Rotterdam. The samples exclude municipalities that have never had cannabis retailers or in which retailers were closed for another reason. 960Robust standard errors are clustered by municipality. Vertical lines around the point estimates represent 90% and 95% confidence intervals, respectively. 961118962