1# RECORD 7f45f70c2c2271b7539c0915321311343666b3842# TITLE: The Grass Isn’t Always Greener: How Crime Influences Property Values Near Urban Greenlines3# YEAR: 2026.0 VENUE: Journal of real estate finance and economics DOI: 10.1007/s11146-025-10026-74# AUTHORS: Jason De Freitas; Fatemeh Kamkar; Mark A. Sunderman; Velma Zahirovic-Herbert5# PAGES: 37 CANDIDATE TABLE PAGES (1-based): [12, 13, 20, 22, 23, 30, 31, 32, 33]6# ABSTRACT: 7 Green spaces, including greenlines and greenways, are widely regarded as amenities that enhance residential property values through aesthetic appeal, recreational opportunities, and improved community connectivity. However, these benefits may be offset by crime, which deters homebuyers and diminishes perceived safety. This study examines how proximity to greenlines affects property values in Memphis, Shelby County, Tennessee, and how this relationship is shaped by local crime conditions. Using a hedonic pricing model and a spatial Difference-in-Differences framework, we analyze property transactions alongside detailed measures of property, violent, and other crime types. We find that while greenlines are generally associated with increased property values, this effect is significantly moderated by crime—especially property crime. In high-crime neighborhoods, proximity to greenlines can decrease property values, suggesting that perceived safety concerns outweigh amenity benefits. Our findings suggest that perceived safety plays a central role in shaping the value residents place on green infrastructure. Although focused on Shelby County, the results offer broader insight into how crime can condition the amenity value of green infrastructure in other urban and suburban contexts.8910===== PAGE 1 (score 9, numeric-paren 0, table-words 0, var-words 9) =====1112Vol.:(0123456789)13The Journal of Real Estate Finance and Economics14https://doi.org/10.1007/s11146-025-10026-715The Grass Isn’t Always Greener: How Crime Influences 16Property Values Near Urban Greenlines17Jason De Freitas1 · Fatemeh Kamkar1 · Mark Sunderman1 · 18Velma Zahirovic‑Herbert1 19Accepted: 24 July 2025 20© The Author(s) 202621Abstract22Green spaces, including greenlines and greenways, are widely regarded as ameni-23ties that enhance residential property values through aesthetic appeal, recreational 24opportunities, and improved community connectivity. However, these benefits may 25be offset by crime, which deters homebuyers and diminishes perceived safety. This 26study examines how proximity to greenlines affects property values in Memphis, 27Shelby County, Tennessee, and how this relationship is shaped by local crime condi-28tions. Using a hedonic pricing model and a spatial Difference-in-Differences frame-29work, we analyze property transactions alongside detailed measures of property, 30violent, and other crime types. We find that while greenlines are generally associ-31ated with increased property values, this effect is significantly moderated by crime—32especially property crime. In high-crime neighborhoods, proximity to greenlines can 33decrease property values, suggesting that perceived safety concerns outweigh amen-34ity benefits. Our findings suggest that perceived safety plays a central role in shap-35ing the value residents place on green infrastructure. Although focused on Shelby 36County, the results offer broader insight into how crime can condition the amenity 37value of green infrastructure in other urban and suburban contexts.38Keywords Greenline · Property values · Crime · Neighborhood · Spatial 39externalities · Spatial difference-in-differences40 * Jason De Freitas 41 42jdefrets@memphis.edu43 44Fatemeh Kamkar 45 46fkamkar@memphis.edu47 48Mark Sunderman 49 50msndrman@memphis.edu51 52Velma Zahirovic‑Herbert 53 54velma.zahirovic-herbert@memphis.edu551 56Department of Finance, Insurance, and Real Estate, Fogelman College of Business & 57Economics, University of Memphis, Memphis, TN 38152, USA585960===== PAGE 12 (score 92, numeric-paren 59, table-words 3, var-words 16) =====6162 63J. De Freitas et al.64Table 3 Summary statistics for sales transactions and housing characteristics by greenline proximity, 652006–202266This table presents summary statistics for property sale prices and housing characteristics for the full 67sample and for properties located within three proximity bands (RINGs) surrounding the greenlines. 68RING1 includes properties within 0.25 miles of a greenline, RING2 includes properties between 0.26 69and 0.50 miles, and RING3 includes properties between 0.51 and 1 mile. Housing and transaction data 70are from the Memphis Area Association of Realtors (MAAR) MLS database. Proximity to greenlines is 71measured using spatial indicators derived from U.S. Census Bureau GIS data (Property transactions were 72geocoded to determine their distance from each greenline, which was used to construct proximity indica-73tors (RING1, RING2, RING3). Each transaction was also matched to a census tract using its geocoded 74location, allowing us to attach annual tract-level crime data. Census tract fixed effects are included in all 75regressions to control for time-invariant neighborhood characteristics). The final three columns report 76t-test results comparing each RING subgroup to the full sample. Statistical significance is denoted by 77***, **, and * at the 1%, 5%, and 10% levels, respectively78b The “t-test mean difference” columns report differences derived from two-sample t-tests comparing 79each RING group to the full sample mean, using unequal variances. These are not simple subtractions 80of meansbut reflect statistical differences accounting for pooled variance. Minor discrepancies between 81reported means and t-test values may occur due to rounding or standard error adjustments82t-test mean differenceb83Variable84Full sample mean85(SD)86RING 1 mean87(SD)88RING 2 mean89(SD)90RING 3 mean91(SD)92Full sample 93vs. ring 194Full sample 95vs. Rin g 296Full sample vs. 97RING 398Sale price99220,443.800100(178,639.500)101217,284.400102(188,493.200)103214,866.800104(180,460.000)105227,297.700106(167,665.000)1074817.586*1087683.102***109−11,085.160***110Acreage1110.290112(0.450)1130.278114(0.365)1150.298116(0.562)1170.295118(0.429)1190.018***120−0.011121−0.008122Square feet1232033.153124(997.223)1252050.939126(1063.973)1272034.957128(999.278)1292015.821130(931.207)131−27.119*132−2.48513328.032*134Property age13540.902136(25.349)13739.696138(26.444)13937.738140(24.761)14144.260142(24.351)1431.838***1444.359***145−5.432***146Year built1471973.903148(24.705)1491975.141150(25.889)1511976.855152(23.973)1531970.668154(23.749)155−1.887 ***156−4.067***1575.233***158Bedrooms1593.238160(0.731)1613.273162(0.730)1633.260164(0.723)1653.191166(0.735)167−0.053***168−0.030**1690.076***170Full baths1712.084172(0.729)1732.125174(0.735)1752.108176(0.704)1772.031178(0.738)179−0.062***180−0.032***1810.086***182Half baths1830.335184(0.515)1850.337186(0.522)1870.379188(0.539)1890.302190(0.490)191−0.003192−0.061***1930.054***194Rooms1956.963196(1.673)1976.949198(1.747)1996.967200(1.699)2016.972202(1.583)2030.021204−0.006205−0.015206Fireplaces2070.937208(0.587)2090.983210(0.601)2110.948212(0.576)2130.887214(0.579)215−0.070***216−0.016*2170.080***218Obs21918,41122063372215046222702822318,41122418,41122518,411226and crime distributions across greenline proximities. These bands standardize meas-227urement zones across greenlines and allow for consistent visual comparisons. Full-228resolution figures enable clearer visualization of census tract boundaries and local 229neighborhood context. Figure 1 illustrates property sale densities near the greenlines 230within 0.25 miles, 0.50 miles and 1 mile from 2006 to 2022. The areas around the 231SFG and SFGE show a higher density of sales while the area around the WR shows 232a lower density of sales for the period.233234235===== PAGE 13 (score 36, numeric-paren 21, table-words 3, var-words 6) =====236237The Grass Isn’t Always Greener: How Crime Influences Property…238Table 4 Summary statistics for crime densities by proximity to greenlines, 2006–2022239This table reports summary statistics for log-transformed crime density measures—violent, property, and other crime—for properties located within three proximity bands 240(RINGs) around greenlines. RING1 includes properties within 0.25 miles of the greenlines, RING2 covers those between 0.26 and 0.50 miles, and RING3 includes prop-241erties between 0.51 and 1 mile. Crime density is defined as the number of crimes per square mile at the census tract level. Crime data are obtained from the Memphis Data 242Hub, while greenline proximity measures are constructed using GIS and spatial indicators derived from U.S. Census Bureau data (Property transactions were geocoded to 243determine their distance from each greenline, which was used to construct proximity indicators (RING1, RING2, RING3). Each transaction was also matched to a census 244tract using its geocoded location, allowing us to attach annual tract-level crime data. Census tract fixed effects are included in all regressions to control for time-invariant 245neighborhood characteristics). The final three columns report t-test results comparing the means of each RING group to the full sample. Statistical significance is denoted 246by ***, **, and * at the 1%, 5%, and 10% levels, respectively247b The “t-test mean difference” columns report differences derived from two-sample t-tests comparing each RING group to the full sample mean, using unequal variances. 248These are not simple subtractions of means but reflect statistical differences accounting for pooled variance. Minor discrepancies between reported means and t-test values 249may occur due to rounding or standard error adjustments250t-test mean differenceb251Variable252Full Sample Mean (SD)253RING 1 Mean (SD)254RING 2 Mean (SD)255RING 3 Mean (SD)256Full Sample 257vs. RING 1258Full Sample 259vs. RING 2260Full Sample vs. RING 3261Violent crime density2624.246 (1.915)2633.528 (1.870)2644.090 (1.929)2655.006 (1.649)2661.095***2670.215***268−1.229***269Property crime density2705.121 (2.051)2714.367 (2.072)2724.942 (2.099)2735.929 (1.673)2741.150***2750.247***276−1.307***277Other crime density2783.382 (1.806)2792.780 (1.690)2803.244 (1.791)2814.023 (1.708)2820.917***2830.189***284−1.037***285Obs28618,41128763372885046289702829018,41129118,41129218,411293294295===== PAGE 14 (score 8, numeric-paren 0, table-words 0, var-words 8) =====296297 298J. De Freitas et al.299Figure 1 shows the density of residential property sales across three standardized 300proximity bands—within 0.25 miles (RING1), 0.26–0.50 miles (RING2), and 0.51–1 301mile (RING3)—of the Shelby Farms Greenline (SFG), its western extension (SFGE), 302and the Wolf River Greenline (WR). We use proximity-based rectangles rather than 303individual points to enable consistent comparison across time and space, and to align 304with the structure of our empirical analysis. Apparent concentrations of sales near 305certain greenline segments reflect underlying patterns in housing supply and neigh-306borhood development. Our regression results control for these spatial and temporal 307differences, allowing for more accurate identification of greenline effects on property 308values.309We collected crime data from the Memphis Police Department and the FBI, 310encompassing information on crime type, location, and timing. Following the crime 311categorization in the economics literature (Lacoe et al., 2018 and Palmer et al., 3122019), the real estate literature (Harris et al., 2018) and the Uniform Crime Statistics 313guidelines from the FBI’s Uniform Crime Reporting (UCR) program, we aggregate 314our crime types into three specific categories: violent crime, property crime, and 315other crime. Violent crime includes homicide, robbery, assault, and kidnapping. In 316contrast, property crime includes property crime, breaking and entering, theft, arson, 317theft of vehicle, and theft from vehicle, and other crimes include weapons offense, 318family offense, traffic, disorder, vehicle recovery, drug-related offences, and liquor 319violations. Each property was matched to crime data corresponding to its census 320tract and year of sale, using tract boundaries defined by the U.S. Census Bureau’s 3212020 American Community Survey (ACS), the most recent delineations available at 322Fig. 1 Property sales density near the greenlines 2006–2022323324325===== PAGE 20 (score 34, numeric-paren 0, table-words 5, var-words 19) =====326327 328J. De Freitas et al.329that crime is a significant consideration in these neighborhoods and may impact 330property values. Although the RING and greenline variables are not explicitly 331reported in the table, their inclusion as binary indicators in the analysis is essen-332tial for examining how proximity to Greenlines might influence property values 333and crime rates. Overall, our summary statistics suggest that both proximity to 334greenlines and neighborhood crime rates play a role in shaping property values.335Table 5 Summary statistics of sales transactions, housing characteristics, and crime, 2006–2022336This table provides descriptive statistics for the key variables used in the analysis, covering the period 337from 2006 to 2022. Variables include property values, housing characteristics, neighborhood features, 338and local crime measures. Sale price, housing and neighborhood characteristics, and year of sale are 339sourced from MAAR MLS data. Crime density variables are calculated at the census tract level, using 3402020 tract definitions from the U.S. Census Bureau’s American Community Survey (ACS), and reflect 341average counts per square mile. Greenline proximity indicators, not shown here, are created using GIS-342based spatial analysis (Property transactions were geocoded to determine their distance from each green-343line, which was used to construct proximity indicators (RING1, RING2, RING3). Each transaction was 344also matched to a census tract using its geocoded location, allowing us to attach annual tract-level crime 345data. Census tract fixed effects are included in all regressions to control for time-invariant neighborhood 346characteristics). For each variable, the table reports the mean, standard deviation, minimum, and maxi-347mum values. All monetary values are in nominal U.S. dollars. The dependent variable used in regression 348analysis, LNSP, is the natural logarithm of the sale price349Variable350Mean351Standard deviation352Minimum353Maximum354LNSP35512.0883560.63535710.31235814.509359Sale price360220,443.800361178,639.50036230,1003632,000,000364Bedrooms3653.2383660.73136713687369Full baths3702.0843710.72937213737374Half baths3750.3353760.51537703785379Rooms3806.9633811.673382338316384Fireplaces3850.9373860.58738703885389Acreage3900.2903910.4503920.0503938.450394Square feet3952033.153396997.22339757239810,689399Property age40040.90240125.3494020403134404Year built4051973.90340624.70540718754082022409Townhouse4100.0354110.18341204131414Condominium4150.0064160.07941704181419Subdivision4200.1534210.36142204231424Neighborhood4250.7194260.45042704281429No pool4300.9154310.28043204331434In-ground pool4350.0694360.25443704381439Above ground pool4400.0064410.07744204431444Carport4450.1634460.36944704481449Covered garage4500.5294510.49945204531454Violent crime density4554.2464561.91545704587.085459Property crime density4605.1214612.05146204638.139464Other crime density4653.3824661.80646704686.390469Obs47018,41147118,41147218,41147318,411474475476===== PAGE 21 (score 23, numeric-paren 0, table-words 3, var-words 17) =====477478The Grass Isn’t Always Greener: How Crime Influences Property…4799 The coefficient results for all variables, including housing and neighborhood characteristics, are pro-480vided in the appendix in Table 8.481Results482The results for our hedonic regression models provide insight into the relationship 483between proximity to greenlines, crime, and property values. Using data on approxi-484mately 18,411 property transactions located within one mile of the Shelby Farms 485Greenline (SFG), Shelby Farms Greenline Extension (SFGE), and Wolf River 486Greenway (WR) between 2006 and 2022, we estimate the effects of greenline prox-487imity and crime rates on property values. Our analysis incorporates fixed effects at 488the year-by-census-tract level to control for unobserved heterogeneity and spatial 489differences.490Hedonic Regression with Difference‑in‑Differences (DiD) Approach Results491Table 6 presents the results of our baseline Difference-in-Differences (DiD) model, 492which extends the standard hedonic regression framework by incorporating interac-493tion terms to evaluate how proximity to greenlines affects property values before 494and after their respective openings.495These models estimate the relationship between greenline proximity, crime 496rates, and property values while controlling for property characteristics, neighbor-497hood characteristics, census tracts, and year fixed effects. The property and neigh-498borhood controls include the number of rooms, bedrooms, full and half bathrooms, 499fireplaces, acreage, square footage, property age, year built, presence of carports or 500garages, pool characteristics, property type (townhouse, condominium), subdivision, 501and neighborhood associations. However, for brevity, the coefficients for these con-502trols are not reported.9503The coefficients on RING1 and RING2 measure the general impact of proxim-504ity to greenlines on property values, independent of time. The results show a posi-505tive effect for properties near SFG, with values increasing by 11.0% in RING1 and 50612.6% in RING2, both statistically significant at the 1% level. The WR, however, 507experiences a more modest decline of 0.69%, which is only marginally significant.508The coefficients on RING1 and RING2 measure the general impact of proxim-509ity to greenlines on property values, independent of time. The results show a posi-510tive effect for properties near SFG, with values increasing by 11.0% in RING1 and 51112.6% in RING2, both statistically significant. This suggests that properties close to 512SFG benefited from greenline proximity before the treatment period. However, for 513WR and SFGE, no significant pre-treatment price premium is observed.514The RING × Post interaction terms capture how the post-treatment period modi-515fies the effect of greenline proximity on property values. In the SFG model, proper-516ties in RING1 experience an 8.2% decrease in value post-treatment, significant at the 51710% level, while properties in RING2 see a 9.9% decline, significant at the 1% level. 518These findings indicate that while proximity to SFG initially adds value, this effect 519diminished post-opening, possibly due to increased foot traffic, congestion, or crime 520521522===== PAGE 22 (score 43, numeric-paren 20, table-words 8, var-words 3) =====523524 525J. De Freitas et al.526concerns. In contrast, WR and SFGE do not exhibit statistically significant changes 527in property values for properties in RING1 or RING2 after the greenline openings.528Overall, these results suggest that the effects of greenlines on property values 529are nuanced and vary by location. While the SFG initially boosts property values, 530this effect declines after its opening. The WR shows a relatively stable trend, while 531SFGE exhibits a strong negative post-treatment effect. These findings highlight the 532importance of considering both spatial and temporal dynamics when evaluating 533greenline impacts on housing markets. Table 6 establishes the foundational proxim-534ity effects that are analyzed further in Tables 7, 8 and 9. The models in these later 535Table 6 Baseline difference-in-differences regression results, 2006–2022 for the greenline openings536This table shows the results of baseline Difference-in-Differences (DiD) regressions estimating the 537impact of greenline openings on property values between 2006 and 2022. The dependent variable is the 538natural log of sale price (LNSP). The models include fixed effects for census tracts (based on 2020 U.S. 539Census data) and year of sale (Property transactions were geocoded to determine their distance from each 540greenline, which was used to construct proximity indicators (RING1, RING2, RING3). Each transaction 541was also matched to a census tract using its geocoded location, allowing us to attach annual tract-level 542crime data. Census tract fixed effects are included in all regressions to control for time-invariant neigh-543borhood characteristics. Stata automatically omits the 2022 year fixed effect to avoid perfect multicol-544linearity. As a result, the Post indicator—equal to 1 for post-greenline-opening transactions—was not 545dropped and is included in the model by default. However, because Post is functionally redundant with 546the full set of year fixed effects, its coefficient does not represent a standalone treatment effect and is 547therefore not reported or interpreted in the table. All treatment effects are identified through the interac-548tion terms). Property and neighborhood characteristics from MLS data are controlled for but not dis-549played for brevity. Proximity to greenlines is captured using geocoded spatial buffers: RING1 includes 550properties within 0.25 miles, RING2 between 0.26 and 0.50 miles, and RING3 (0.51–1 mile) serves as 551the omitted baseline group. Treatment effects are identified through the interaction terms (RING1 × Post 552and RING2 × Post), which test for differential effects by proximity to greenlines after each greenline’s 553respective opening. Separate results are shown for the full sample and for properties near the Wolf River 554Greenline (WR), Shelby Farms Greenline (SFG), and Shelby Farms Greenway Extension (SFGE). Statis-555tical significance is denoted by ***, **, and * at the 1%, 5%, and 10% levels, respectively556Dependent variable: LNSP557Variable558Full sample mean 559(SD)560WR mean (SD)561SFG mean (SD)562SFGE mean (SD)563RING15640.006565−0.0085660.110**567−0.030568(0.035)569(0.043)570(0.040)571(0.028)572RING1 x Post573−0.005574−0.002575−0.082*5760.019577(0.036)578(0.041)579(0.042)580(0.032)581RING25820.031583−0.0295840.126***585−0.018586(0.027)587(0.039)588(0.025)589(0.010)590RING2 x Post591−0.0145920.085593−0.099***5940.014595(0.028)596(0.045)597(0.029)598(0.016)599Property controls600Yes601Yes602Yes603Yes604Neighborhood controls605Yes606Yes607Yes608Yes609Year fixed effects610Yes611Yes612Yes613Yes614Census tract fixed effects615Yes616Yes617Yes618Yes619R-squared6200.8416210.9046220.8306230.855624Obs62518,411626180662711,4386285167629630631===== PAGE 23 (score 87, numeric-paren 57, table-words 9, var-words 5) =====632633The Grass Isn’t Always Greener: How Crime Influences Property…634Table 7 Extended difference-in-differences regression results: property crime, pre- and post-2010 and 6352016 for the greenline openings636This table presents results from extended Difference-in-Differences (DiD) regression models estimating the 637effect of property crime density on housing prices near greenlines from 2006 to 2022. The models extend the 638baseline specification by incorporating property crime measures and their interactions with greenline proximity 639and post-treatment periods (2010 and 2016). The dependent variable is the natural log of sale price (LNSP). 640Treatment effects are identified through interaction terms (RING1 × Post, RING2 × Post, Log Property Crime 641Density × Post, RING1 × Log Property Crime Density × Post, RING2 × Log Property Crime Density × Post), 642which test for how property crime influences property values across space and time. Separate regressions are 643reported for the full sample and for properties near the Wolf River Greenline (WR), Shelby Farms Greenline 644(SFG), and Shelby Farms Greenway Extension (SFGE). Fixed effects for year of sale and census tracts (2020 645U.S. Census data) are included (Property transactions were geocoded to determine their distance from each 646greenline, which was used to construct proximity indicators (RING1, RING2, RING3). Each transaction was 647also matched to a census tract using its geocoded location, allowing us to attach annual tract-level crime data. 648Census tract fixed effects are included in all regressions to control for time-invariant neighborhood character-649istics. Year fixed effects are also included; Stata automatically omits the 2022 year dummy to avoid perfect 650multicollinearity. As a result, the Post indicator—equal to 1 for post-greenline-opening transactions—was not 651dropped and is included in the model by default. However, because Post is fully determined by the year fixed 652effects, its coefficient is not reported or interpreted in the table. All treatment effects are captured through inter-653action terms involving Post). Structural housing and neighborhood characteristics from MLS data are also con-654trolled for but omitted from the table for brevity. Proximity to greenlines is measured by RING1 (≤ 0.25 miles) 655and RING2 (0.26–0.50 miles), with RING3 (0.51–1 mile) serving as the omitted reference group. Key interac-656tion terms between property crime, greenline proximity, and the post-treatment period allow us to assess how 657property values respond to crime across space and time. Statistical significance is denoted by ***, **, and * for 658the 1%, 5%, and 10% levels, respectively659Dependent variable: LNSP660Variable661Full Sample Mean 662(SD)663WR Mean 664(SD)665SFG Mean 666(SD)667SFGE Mean 668(SD)669RING1670−0.243***6710.164*672−0.235673−0.153*674(0.046)675(0.065)676(0.138)677(0.078)678RING1 x Post6790.179***680−0.159*6810.293*6820.136*683(0.046)684(0.063)685(0.146)686(0.060)687RING2688−0.287***6890.080*690−0.304**691−0.181***692(0.038)693(0.038)694(0.142)695(0.047)696RING2 x Post6970.229***6980.0116990.306**7000.139***701(0.047)702(0.050)703(0.131)704(0.018)705Log Property crime density706−0.057***7070.024708−0.108***7090.002710(0.014)711(0.015)712(0.031)713(0.016)714Log Property crime density x Post7150.042***716−0.029***7170.070***7180.025***719(0.008)720(0.007)721(0.018)722(0.002)723RING1 x Log Property crime density x Post724−0.030***7250.041726−0.047**727−0.024728(0.010)729(0.023)730(0.021)731(0.018)732RING2 x Log Property crime density x Post733−0.044***7340.010735−0.054***736−0.026***737(0.009)738(0.014)739(0.019)740(0.004)741Property controls742Yes743Yes744Yes745Yes746Neighborhood controls747Yes748Yes749Yes750Yes751Year fixed effects752Yes753Yes754Yes755Yes756Census tract fixed effects757Yes758Yes759Yes760Yes761Adj. R-squared7620.8437630.9057640.8317650.856766Obs76718,411768180676911,4387705167771772773===== PAGE 24 (score 16, numeric-paren 0, table-words 5, var-words 6) =====774775 776J. De Freitas et al.777tables explore how the baseline effects evolve post-greenline openings and examine 778the interaction between crime and these amenities. Thus, Table 6 provides a com-779parative context for evaluating shifts in property value dynamic over time.780Difference‑in‑Differences (DiD) Analysis781The DiD regression results, presented in Table 7 (as well as in Tables 8 and 9 782located in the appendix) examine the interaction between proximity to greenlines, 783crime, and the pre- and post-opening periods of the SFG, SFGE, and WR. These 784models allow us to assess how greenline opening influenced property values over 785time, while accounting for crime rate densities.786Table 7 presents the estimated coefficients and standard errors for the interaction terms 787between proximity to greenlines, property crime density, and the pre- and post-opening 788periods. The results provide key insights into how property crime influences housing val-789ues in different greenline areas over time. In the full sample, property crime density has a 790statistically significant negative effect on property values (−5.7% at the 1% level), indicat-791ing that higher crime rates are generally associated with lower home prices. However, this 792effect varies across greenline areas. While WR shows a small, positive but insignificant 793coefficient (2.4%), SFG exhibits the most pronounced negative effect (−10.8% at the 1% 794level), followed by SFGE, which has an insignificant coefficient (0.2%).795The interaction between property crime density and the post-treatment period 796reveals significant variations in how greenlines mediate crime effects. In the full 797sample, the interaction term is positive and significant (4.2% at the 1% level), sug-798gesting that after greenline openings, properties in high-crime areas experienced 799appreciation in value. However, WR shows a significant negative post-opening inter-800action (−2.9% at the 1% level), implying that crime had a worsening effect on prop-801erty values after WR’s opening. In contrast, SFG (7.0% at the 1% level) and SFGE 802(2.5% at the 1% level) exhibit significant positive effects, indicating that these green-803lines may have enhanced property values despite crime.804The key findings emerge from the triple interaction terms, which capture how 805crime influences property values near greenlines post-opening. In the full sample, 806the RING1 x Property Crime Density x Post interaction is negative and significant 807(−3.0% at the 1% level), implying that properties closest to greenlines in high-crime 808areas depreciated post-opening. This effect is particularly strong for SFG (−4.7% 809at the 5% level), suggesting that crime may have diminished the perceived amenity 810value of the greenline. WR and SFGE, however, show insignificant effects for this 811interaction, indicating less clear post-opening crime impacts in these areas.812For RING2, property crime density has a strong positive effect in the full sample 813(5.9% at the 1% level), suggesting that properties in the 0.26 to 0.50-mile range appreci-814ated in higher-crime areas. However, WR has a significant negative effect (−2.2% at the 81510% level), indicating crime’s more detrimental role in this region. The post-treatment 816interaction (RING2 x Property Crime Density x Post) is negative and significant in the 817full sample (−4.4% at the 1% level), implying that properties in this range were nega-818tively affected by crime after greenline openings. This effect is particularly strong for 819820821===== PAGE 30 (score 251, numeric-paren 209, table-words 2, var-words 14) =====822823 824J. De Freitas et al.825Table 8 Baseline hedonic regression models of property values near greenlines for 2006 to 2022826Dependent variable: LNSP827Variable828(1)829(2)830(3)831(4)832(5)833(6)834Rooms8350.012***8360.012***8370.012***8380.012***8390.012***8400.012***841(0.004)842(0.004)843(0.004)844(0.004)845(0.004)846(0.004)847Bedrooms8480.0068490.0068500.0068510.0068520.0068530.007854(0.011)855(0.011)856(0.011)857(0.011)858(0.011)859(0.011)860Full baths8610.125***8620.125***8630.125***8640.125***8650.124***8660.124***867(0.021)868(0.021)869(0.021)870(0.021)871(0.021)872(0.021)873Half baths8740.034**8750.033**8760.033**8770.033**8780.033**8790.033**880(0.015)881(0.015)882(0.015)883(0.015)884(0.015)885(0.015)886Fireplaces8870.086***8880.086***8890.086***8900.086***8910.086***8920.086***893(0.012)894(0.012)895(0.012)896(0.012)897(0.012)898(0.012)899Acreage9000.0079010.0069020.0079030.0079040.0079050.007906(0.019)907(0.019)908(0.019)909(0.019)910(0.019)911(0.019)912Square feet9130.000***9140.000***9150.000***9160.000***9170.000***9180.000***919(0.000)920(0.000)921(0.000)922(0.000)923(0.000)9240.000925Property age9260.039***9270.039***9280.039***9290.039***9300.042***9310.042***932(0.001)933(0.001)934(0.001)935(0.001)936(0.001)937(0.001)938Year built9390.042***9400.042***9410.042***9420.042***9430.044***9440.044***945(0.001)946(0.001)947(0.001)948(0.001)949(0.002)950(0.001)951Carport9520.066***9530.066***9540.066***9550.066***9560.066***9570.065***958(0.017)959(0.017)960(0.017)961(0.017)962(0.017)963(0.017)964Covered garage9650.074***9660.073***9670.073***9680.073***9690.073***9700.073***971(0.008)972(0.008)973(0.008)974(0.008)975(0.008)976(0.008)977No pool9780.088**9790.089**9800.089**9810.089**9820.088**9830.089**984(0.040)985(0.040)986(0.040)987(0.040)988(0.040)989(0.040)990In-ground pool9910.105**9920.106**9930.106**9940.106**9950.106**9960.107**997(0.040)998(0.039)999(0.040)1000(0.040)1001(0.039)1002(0.040)1003Above ground pool10040.124***10050.122***10060.122***10070.123***10080.122***10090.122***1010(0.044)1011(0.044)1012(0.044)1013(0.044)1014(0.044)1015(0.044)1016Townhouse1017−0.213*** −0.210*** −0.210*** −0.210*** −0.210*** −0.210***1018(0.022)1019(0.022)1020(0.022)1021(0.022)1022(0.022)1023(0.022)1024Condominium1025−0.458**1026−0.457**1027−0.456**1028−0.457**1029−0.458**1030−0.458**1031(0.181)1032(0.181)1033(0.180)1034(0.180)1035(0.180)1036(0.180)1037Subdivision10380.01010390.01010400.01010410.00910420.00910430.0091044(0.014)1045(0.014)1046(0.014)1047(0.014)1048(0.014)1049(0.014)1050Neighborhood1051−0.041*1052−0.040*1053−0.040*1054−0.040*1055−0.040*1056−0.040*1057(0.058)1058(0.021)1059(0.021)1060(0.021)1061(0.021)1062(0.021)1063RING110640.00210650.00210660.00210670.00310680.00210690.0021070(0.021)1071(0.021)1072(0.021)1073(0.021)1074(0.021)1075(0.021)1076RING210770.02110780.02110790.02110800.02010810.02010820.0201083(0.014)1084(0.014)1085(0.014)1086(0.014)1087(0.014)1088(0.014)1089SFG10900.00610910.00710920.00610930.00510940.005109510961097===== PAGE 31 (score 65, numeric-paren 18, table-words 14, var-words 15) =====10981099The Grass Isn’t Always Greener: How Crime Influences Property…1100This table presents results from hedonic regressions estimating the relationship between property values 1101and proximity to greenlines, housing characteristics, and neighborhood features (Property transactions 1102were geocoded to determine their distance from each greenline, which was used to construct proximity 1103indicators (RING1, RING2, RING3). Each transaction was also matched to a census tract using its geo-1104coded location, allowing us to attach annual tract-level crime data. Census tract fixed effects are included 1105in all regressions to control for time-invariant neighborhood characteristics). The dependent variable is 1106the log of sale price (LNSP). Models include structural attributes, subdivision and property type indica-1107tors, and fixed effects for year of sale and census tracts. Proximity to greenlines is represented by RING1 1108(≤ 0.25 miles) and RING2 (0.26–0.50 miles), with RING3 (0.51–1 mile) as the reference. Greenline indi-1109cators include SFG and WR; SFGE is excluded and serves as the base group. Crime variables (violent, 1110property, and other) are added in columns (4) through (6). These baseline models inform the Difference-1111in-Differences analysis presented in Table 6. Statistical significance is indicated by ***, **, and * at the 11121%, 5%, and 10% levels, respectively1113Table 8 (continued)1114Dependent variable: LNSP1115Variable1116(1)1117(2)1118(3)1119(4)1120(5)1121(6)1122(0.017)1123(0.017)1124(0.017)1125(0.017)1126(0.018)1127WR11280.04811290.04811300.04811310.04711320.0481133(0.037)1134(0.038)1135(0.037)1136(0.037)1137(0.038)1138Log Violent crime density11390.00711400.0261141(0.028)1142(0.027)1143Log Property crime 1144density1145−0.0061146−0.0041147(0.015)1148(0.013)1149Log Other crime density1150−0.040**1151−0.047***1152(0.017)1153(0.017)1154Year fixed effects1155Yes1156Yes1157Yes1158Yes1159Yes1160Yes1161Census tract fixed effects1162Yes1163Yes1164Yes1165Yes1166Yes1167Yes1168Adj. R-squared11690.84011700.84111710.84111720.84111730.84111740.8421175Obs117618,411117718,411117818,411117918,411118018,411118118,41111820.26–0.50 mi, and 0.51–1 mi) from the three greenlines (SFG, WR, SFGE). Vertical 1183lines indicate the opening years of each greenline.1184Table 8 presents the housing and neighborhood characteristics from the hedonic 1185models estimated in Table 6, which were not reported there for brevity. This table 1186includes coefficients for additional variables related to housing features, neighbor-1187hood controls, RINGs, the greenlines and the various crimes reported in this study. 1188The table spans the 2006–2022 period and includes variables such as the number of 1189rooms, bedrooms, full and half baths, fireplaces, and property characteristics like 1190acreage, square footage, and property age. The greenline indicators (SFG and WR) 1191and RINGs (with RING3 as the baseline) are also included, as well as crime vari-1192ables for violent, property, and other crimes. All other results remain consistent with 1193those in Table 6. This supplemental table provides a detailed view of the estimated 1194coefficients for the housing and neighborhood characteristics used in the analysis.1195Table 9 presents the estimated coefficients and standard errors for the interac-1196tion terms between proximity to greenlines, violent crime densities, and the pre- 1197and post-opening periods. The results highlight significant triple interaction effects, 119811991200===== PAGE 32 (score 81, numeric-paren 52, table-words 8, var-words 6) =====12011202 1203J. De Freitas et al.1204Table 9 Extended difference-in-differences regression results: violent crime, pre- and post-2010 and 12052016 for the greenline openings1206This table presents results from extended Difference-in-Differences (DiD) regression models examining 1207the relationship between violent crime density and housing prices near greenlines from 2006 to 2022. 1208These models build on the baseline hedonic specification by incorporating interactions between violent 1209crime density, greenline proximity, and post-treatment periods (post-2010 and post-2016). The depend-1210ent variable is the natural log of sale price (LNSP). Treatment effects are identified through interaction 1211terms (RING1 × Post, RING2 × Post, Log Violent Crime Density × Post, RING1 × Log Violent Crime 1212Density × Post, RING2 × Log Violent Crime Density × Post), which test for how violent crime influences 1213property values across space and time. Analyses are conducted for the full sample and separately by 1214greenline: the Wolf River Greenline (WR), the Shelby Farms Greenline (SFG), and the Shelby Farms 1215Greenway Extension (SFGE). Fixed effects for year of sale and census tracts (based on 2020 U.S. Census 1216data) are included (Property transactions were geocoded to determine their distance from each greenline, 1217which was used to construct proximity indicators (RING1, RING2, RING3). Each transaction was also 1218matched to a census tract using its geocoded location, allowing us to attach annual tract-level crime data. 1219Census tract fixed effects are included in all regressions to control for time-invariant neighborhood char-1220acteristics. Year fixed effects are also included; Stata automatically omits the 2022 year dummy to avoid 1221perfect multicollinearity. As a result, the Post indicator—equal to 1 for post-greenline-opening transac-1222tions—was not dropped and is included in the model by default. However, because Post is fully deter-1223mined by the year fixed effects, its coefficient is not reported or interpreted in the table. All treatment 1224effects are captured through interaction terms involving Post.). Structural property features and neighbor-1225hood characteristics from MLS data are also controlled for, but coefficients are omitted from the table for 1226brevity. Greenline proximity is captured using RING1 (≤ 0.25 miles) and RING2 (0.26–0.50 miles), with 1227RING3 (0.51–1 mile) serving as the omitted reference category. Key interaction terms measure how vio-1228lent crime density affects property values in close proximity to greenlines before and after their openings. 1229Statistical significance is indicated at the 1%, 5%, and 10% levels by ***, **, and *, respectively1230Dependent variable: LNSP1231Variable1232Full Sample Mean 1233(SD)1234WR Mean 1235(SD)1236SFG Mean 1237(SD)1238SFGE Mean (SD)1239RING11240−0.204***12410.1511242−0.1561243−0.146*1244(0.034)1245(0.094)1246(0.147)1247(0.075)1248RING1 x Post12490.153***1250−0.16312510.253*12520.128*1253(0.047)1254(0.097)1255(0.138)1256(0.055)1257RING21258−0.234***12590.0821260−0.2181261−0.145**1262(0.027)1263(0.047)1264(0.143)1265(0.047)1266RING2 x Post12670.184***1268−0.00812690.252*12700.099***1271(0.045)1272(0.043)1273(0.128)1274(0.020)1275Log Violent crime density1276−0.044*12770.0251278−0.01412790.0081280(0.025)1281(0.014)1282(0.038)1283(0.026)1284Log Violent crime density x Post12850.041***1286−0.044***12870.062***12880.027***1289(0.008)1290(0.007)1291(0.019)1292(0.002)1293RING1 x Log Violent crime density x Post1294−0.033***12950.0541296−0.051**1297−0.0301298(0.012)1299(0.049)1300(0.024)1301(0.021)1302RING2 x Log Violent crime density x Post1303−0.046***13040.0201305−0.056**1306−0.025***1307(0.010)1308(0.012)1309(0.021)1310(0.006)1311Property controls1312Yes1313Yes1314Yes1315Yes1316Neighborhood controls1317Yes1318Yes1319Yes1320Yes1321Year fixed effects1322Yes1323Yes1324Yes1325Yes1326Census tract fixed effects1327Yes1328Yes1329Yes1330Yes1331Adj. R-squared13320.84313330.90513340.83113350.8561336Obs133718,41113381806133911,43813405167134113421343===== PAGE 33 (score 82, numeric-paren 52, table-words 9, var-words 5) =====13441345The Grass Isn’t Always Greener: How Crime Influences Property…1346Table 10 Extended difference-in-differences regression results: other crime, pre- and post-2010 and 13472016 for the greenline openings1348Dependent variable: LNSP1349Variable1350Full Sample 1351Mean (SD)1352WR Mean (SD)1353SFG Mean (SD)1354SFGE Mean (SD)1355RING11356-0.123***13570.0541358-0.1441359-0.0701360(0.021)1361(0.039)1362(0.113)1363(0.050)1364RING1 x Post13650.064**1366-0.04713670.179**13680.0491369(0.027)1370(0.030)1371(0.089)1372(0.047)1373RING21374-0.145***13750.0061376-0.1931377-0.086*1378(0.019)1379(0.057)1380(0.112)1381(0.039)1382RING2 x Post13830.089***13840.05413850.167*13860.070***1387(0.026)1388(0.058)1389(0.083)1390(0.010)1391Log Other crime density1392-0.075***13930.0161394-0.092***1395-0.039*1396(0.014)1397(0.021)1398(0.032)1399(0.018)1400Log Other crime density x Post14010.034***1402-0.041***14030.063***14040.028**1405(0.007)1406(0.008)1407(0.015)1408(0.009)1409RING1 x Log Other crime density x 1410Post1411-0.020*14120.0161413-0.042**1414-0.0111415(0.011)1416(0.028)1417(0.019)1418(0.031)1419RING2 x Log Other crime density x 1420Post1421-0.036***14220.0011423-0.049**1424-0.025**1425(0.009)1426(0.024)1427(0.017)1428(0.008)1429Property controls1430Yes1431Yes1432Yes1433Yes1434Neighborhood controls1435Yes1436Yes1437Yes1438Yes1439Year fixed effects1440Yes1441Yes1442Yes1443Yes1444Census tract fixed effects1445Yes1446Yes1447Yes1448Yes1449Adj. R-squared14500.84314510.90514520.83114530.8561454Obs145518,41114561806145711,438145851671459This table results from extended Difference-in-Differences (DiD) regression models evaluating the relationship between 1460other crime density and housing prices near greenlines from 2006 to 2022. These models include interaction terms 1461between other crime density, greenline proximity, and post-treatment periods (post-2010 and post-2016) to assess how 1462property values respond to less severe forms of crime over time. The dependent variable is the natural log of sale price 1463(LNSP). Treatment effects are identified through interaction terms (RING1 × Post, RING2 × Post, Log Other Crime Den-1464sity × Post, RING1 × Log Other Crime Density × Post, RING2 × Log Other Crime Density × Post), which test for how 1465other crime influences property values across space and time. Separate regressions are estimated for the full sample as 1466well as for properties near the Wolf River Greenline (WR), Shelby Farms Greenline (SFG), and Shelby Farms Green-1467way Extension (SFGE). Fixed effects for year of sale and census tracts (based on 2020 U.S. Census data) are included 1468(Property transactions were geocoded to determine their distance from each greenline, which was used to construct 1469proximity indicators (RING1, RING2, RING3). Each transaction was also matched to a census tract using its geocoded 1470location, allowing us to attach annual tract-level crime data. Census tract fixed effects are included in all regressions to 1471control for time-invariant neighborhood characteristics. Year fixed effects are also included; Stata automatically omits 1472the 2022 year dummy to avoid perfect multicollinearity. As a result, the Post indicator—equal to 1 for post-greenline-1473opening transactions—was not dropped and is included in the model by default. However, because Post is fully deter-1474mined by the year fixed effects, its coefficient is not reported or interpreted in the table. All treatment effects are cap-1475tured through interaction terms involving Post.). Structural housing and neighborhood characteristics from MLS data 1476are also controlled for, though omitted from the table for brevity. Greenline proximity is measured using RING1 (≤ 0.25 1477miles) and RING2 (0.26–0.50 miles), with RING3 (0.51–1 mile) serving as the omitted reference group. Key interac-1478tions examine how the effect of other crime density on housing prices varies by proximity and across post-treatment 1479periods. Statistical significance is indicated at the 1%, 5%, and 10% levels by ***, **, and *, respectively148014811482===== PAGE 34 (score 11, numeric-paren 0, table-words 2, var-words 7) =====14831484 1485J. De Freitas et al.1486particularly in RING1 and RING2, where higher violent crime density in post-treat-1487ment periods is associated with diminished capitalization of greenlines. Specifically, 1488the negative coefficient on the triple interaction term (RING1 x Violent crime density 1489x Post) suggests that, in high-crime areas, the greenline effect on property prices is 1490even more pronounced after the greenline’s opening, particularly in RING2. These 1491results reinforce the main analysis findings and the broader relationship between 1492crime and greenline proximity, though property crime is emphasized in the main 1493discussion for its stronger link to homebuyer decision-making.1494Table 10 presents the estimated coefficients and standard errors for the interaction 1495terms between proximity to greenlines, other crime densities, and the pre- and post-1496opening periods. The results show significant triple interaction effects, particularly 1497in RING1 and RING2, where the negative interaction between other crime density 1498and the post-treatment period suggests that areas with higher crime densities experi-1499ence a more pronounced negative effect from greenline proximity. Specifically, the 1500coefficients on the triple interaction terms (RING1 x Other crime density x Post) 1501and (RING2 x Other crime density x Post) indicate that higher other crime density 1502in post-treatment periods dampens the positive capitalization effect of the green-1503lines, particularly in RING2. These findings align with the broader analysis on crime 1504and greenline proximity, further emphasizing the nuanced impact of crime types on 1505property values.1506Authors Contributions All authors contributed to the study’s conception and design. Material prepara-1507tion, data collection, and analysis were performed by J. De Freitas, F. Kamkar, V. Zahirovic-Herbert, and 1508M. Sunderman. The first draft of the manuscript was written by J. De Freitas, and all authors reviewed 1509and provided input on previous versions of the manuscript. All authors read and approved the final 1510manuscript.1511Funding The authors declare that no funds, grants, or other support were received during the preparation 1512of this manuscript.1513Data Availability The data used in this study were obtained under a contract with Memphis Area Asso-1514ciation of Realtors (MAAR) and are not publicly available due to restrictions on sharing. Access to the 1515data is limited by the terms of our agreement with the provider.1516Declarations 1517We, the submitting authors, confirm that this manuscript is an original piece of empirical research, has not 1518been previously published, and is not currently under evaluation by any other journal.1519Ethical Approval Not applicable.1520Consent to Participate Not applicable.1521Consent to Publish All authors give their explicit consent to publish this manuscript.1522Competing interests The authors have no relevant financial or non-financial interests to disclose.1523Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, 1524which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long 1525as you give appropriate credit to the original author(s) and the source, provide a link to the Creative 1526Commons licence, and indicate if changes were made. The images or other third party material in this 1527