The Impact of Recreational Homes on Agricultural Land Use

Charles Towe; Zhenshan Chen

doi:10.3368/le.090420-0140R1

Article Figures & Data

Figures

Tables
Additional Files

Download figure
Open in new tab
Download powerpoint
Figure 1
Example of Land Cover and Parcel Layers
Note: Cropland data are not used as examples, but they have the same resolution as the National Landcover Data.
Download figure
Open in new tab
Download powerpoint
Figure 2
Second-Home Shed, New York City: (a) Distribution of New York City Homeowners in Upstate New York; (b) New York City Homeowners and Commuters
Source: Data are from New York State tax records and the Journey to Work survey.

View popup

Table 3

Difference-in-Difference Matching Results: Average Treatment Effects on the Treated (ATT) for Columbia County with Sample Limited by Acreage

Outcome: Change in Land Cover Proportion, 2001–2012					Treatment: Transactions from Farmers to Nonresidents
	Regression Adjustment with Radius Matched Sample (n=3, r=0.01)			Regression Adjustment with Radius Matched Sample (n=5, r=0.01)			Regression Adjustment with Radius Matched Sample (n=6, r=0.01)
≤ (acres)^{^a}	∞	90	70	∞	90	70	∞	90	70
ATT on Ag^{^b} change	−0.1208^**	−0.1523^*	−0.1530^*	−0.1132^**	−0.1482^*	−0.1250	−0.1168^**	−0.1371^*	−0.1284
ATT on Ag^{^b} change	(0.0578)	(0.0858)	(0.0832)	(0.0570)	(0.0846)	(0.0845)	0(0.0569)	(0.0858)	(0.0850)
ATT on hay change	0.1293^**	0.1431	0.1032	0.1139^*	0.1377	0.0786	0.1124^*	0.1378	0.0843
ATT on hay change	(0.0620)	(0.0894)	(0.0942)	(0.0609)	(0.0894)	(0.0889)	(0.0606)	(0.0911)	(0.0906)
ATT on mismatch change	0.0837	0.0568	0.0514	0.0709	0.0581	0.0282	0.0716	0.0591	0.0341
ATT on mismatch change	(0.0529)	(0.0822)	(0.0816)	(0.0533)	(0.0802)	(0.0775)	(0.0531)	0 (0.0817)	(0.0793)
PS-test (B/R)	41.0/0.91	43.6/0.38	46.7/1.04	27.5/1.13	39.8/0.51	34.0/0.92	23.8/1.00	43.2/0.51	40.3/0.47
N treated: on support (all)	46 (49)	28 (32)	29 (32)	46 (49)	28 (32)	29 (32)	46 (49)	28 (32)	29 (32)
N control: matched (all)	104 (609)	67 (401)	64 (345)	155 (609)	102 (401)	98 (345)	179 (609)	115 (401)	114 (345)
N total	150 (658)	95 (433)	93 (377)	201 (658)	130 (433)	127 (377)	225 (658)	143 (433)	143 (377)

Note: Treatment: transactions from farmer to nonresident happen between 2002 and 2010; control = farmer owner throughout the whole period. Robust standard errors are in parentheses.
↵a Extra-large agricultural parcels are less likely to be recreational second homes, so we can see estimation with large parcels (> 90 acres) tend to give smaller ATTs. When we further restrict the sample to 70 acres (29 treated), the increasing variances dominate the changes.
↵b Ag denotes agricultural land cover, except hay.
↵* p < 0.1;
↵** p < 0.05;
*** p < 0.01.

View popup

Table 1

Summary Statistics for Columbia

Variable	Farmland Transferred to Nonresidents		Farmland Stay Owned by Residents		t-test/Total
Variable	Mean	Std. Dev.	Mean	Std. Dev.	P(\|T\|>\|t\|)
Agricultural ratio before treatment	0.1720	0.2654	0.1468	0.2394	0.4829
Hay ratio before treatment	0.4291	0.2829	0.4740	0.2857	0.2896
Forest before treatment	0.2742	0.2758	0.2509	0.2502	0.5336
Agricultural ratio change	−0.0027	0.3019	0.0913	0.2772	0.0237^**
Hay ratio change	−0.1334	0.3235	−0.2080	0.2840	0.0805^*
Mismatch ratio change	−0.0996	0.2744	−0.1565	0.2586	0.1406
Forest ratio change	0.1478	0.1172	0.1309	0.1346	0.3934
Vacant ratio change	0.0486	0.0639	0.0424	0.0680	0.5349
Wetland ratio change	−0.0650	0.1063	−0.0588	0.1180	0.7196
Developed ratio change	0.0047	0.0266	0.0022	0.0520	0.7426
Acreage	82.97	85.16	74.30	81.75	0.4766
Ln (total assessed value)	11.78	1.321	11.61	1.321	0.3974
Ln (assessed land value)	11.52	1.072	11.32	1.139	0.2316
Stream or pond in parcel	0.4693	0.5042	0.3941	0.4891	0.3013
Divided by road	0.0204	0.1429	0.0805	0.2722	0.1274
Close to primary road	0.4694	0.5042	0.3842	0.4868	0.2405
Ln (distance to Hudson city)	9.657	0.5705	9.764	0.5549	0.1958
Ln (average elevation)	4.723	0.5935	4.621	0.5985	0.2481
Ln (elevation range)	3.457	0.8094	3.128	1.005	0.0262^**
Annual precipitation	41.87	1.028	41.74	1.466	0.5571
Erosion %	0.4437	0.3201	0.3702	0.2983	0.0996^*
Excess water %	0.1922	0.1854	0.2330	0.2433	0.2512
Soil limit %	0.3265	0.3402	0.3637	0.3239	0.4403
Prime farmland %	0.6288	0.2837	0.6468	0.2779	0.6631
No. of observations	49		609		658

↵* p < 0.1;
↵** p < 0.05;
*** p < 0.01.

View popup

Table 2

Difference-in-Difference Matching Results: Average Treatment Effects on Treated (ATT) for Columbia County

Outcome: Change in Land Cover Proportion, 2001–2012				Treatment: Transactions from Farmers to Nonresidents
Estimator	OLS Coefficient	Radius Match 1 (n=3, Caliper=0.01)	Regression Adjustment with Matched Sample 1	Radius Match 2 (n=5, Caliper=0.01)	Regression Adjustment with Matched Sample 2	Radius Match 3 (n=6, Caliper=0.01)	Regression Adjustment with Matched Sample 3
ATT on Ag^{^a} change	−0.0896^**	−0.1258^**	−0.1208^**	−0.1105^**	−0.1132^**	−0.1141^**	−0.1168^**
ATT on Ag^{^a} change	(0.0411)	(0.0637)	(0.0578)	(0.0581)	(0.0570)	(0.0557)	(0.0569)
ATT on hay change	0.0724^*	0.1293^*	0.1293^**	0.1090^*	0.1139^*	0.1075^*	0.1124^*
ATT on hay change	0 (0.0420)	0 (0.0668)	(0.0620)	(0.0633)	(0.0609)	(0.0607)	(0.0606)
ATT on mismatch change	0.0479	0.0842	0.0837	0.0638	0.0709	0.0638	0.0716
ATT on mismatch change	(0.0381)	(0.0573)	(0.0529)	(0.0551)	(0.0533)	0 (0.0518)	(0.0531)
PS-test (B/R)		41.00/.91	41.0/0.91	27.5/1.13	27.5/1.13	23.8/1.00	23.8/1.00
N treated: on support (all)	49	46 (49)	46 (49)	46 (49)	46 (49)	46 (49)	46 (49)
N control: matched (all)	609	104 (609)	104 (609)	155 (609)	155 (609)	179 (609)	179 (609)
N total	658	150 (658)	150 (658)	201 (658)	201 (658)	225 (658)	225 (658)

Note: Treatment: farmland transacted from residents to nonresidents between 2002 and 2010; control = farmland owned by residents throughout the study period. We use propensity matching with the radius set to 0.01. To implement the common support condition, we exclude three observations in the treatment group that do not satisfy that each treated has at least six nearest neighbors in the control group within the radius 0.01. We carry out two kinds of estimators based on matching with a different number of near neighbors (n = 3, 5, 6): radius matching (RM) and regression adjustment with the matched sample (RA). Bootstrapping (2,000 replications for RMs) or robust (for RAs) standard errors are in parentheses.
↵a Ag denotes agricultural land cover, except hay.
↵* p < 0.1;
↵** p < 0.05;
*** p < 0.01.

View popup

Table 4

Placebo Test Results-Average Treatment Effects on the Treated (ATT) for Columbia

Outcome: Change in Land Cover Proportion, 2001–2006			Treatment: Transactions from Farmers to Nonresidents, 2006–2010^{^a}
Estimator^{^a}	OLS coefficient	Regression Adjustment with Matched Sample (n=4, Caliper=0.01)	Regression Adjustment with Matched Sample (n=5, Caliper=0.01)	Regression Adjustment with Matched Sample (n=6, Caliper=0.01)
ATT on Ag^{^b} change	0.0169^*	0.0159	0.0145	0.0140
ATT on Ag^{^b} change	(0.0087)	(0.0146)	(0.0151)	(0.0151)
ATT on hay change	−0.0195^**	−0.0196	−0.0183	−0.0178
ATT on hay change	(.0083)	(0.0136)	(0.0140)	(0.0140)
ATT on mismatch change	−0.0142^**	−0.0141	−0.0127	−0.0127
ATT on mismatch change	(0.0063)	(0.0090)	(0.0094)	(0.0093)
PS-test (B/R)	—	37.5/0.89	30.3/0.94	30.5/0.85
N treated: on support (all)	47	44 (47)	44 (47)	44 (47)
N control: matched (all)	609	142 (609)	167 (609)	191 (609)
N total	656	186 (656)	211 (656)	235 (656)

Note: We do not claim anticipation effects, since we do not get consistently significant placebo test results like Chabé-Ferret and Subervie (2013). Robust standard errors in parentheses.
↵a Control = farmer owner throughout the whole period; we employ propensity matching with the caliper set to 0.01. To implement the common support condition, we exclude three observations in the treatment group, which do not satisfy that each treated has at least six nearest neighbors in the control group within the radius 0.01. We carry out regression adjustment with radius matched samples on different number of near neighbors (N = 4,5,6). All matching estimators give us small insignificant trend differences in the opposite direction of the estimated effects. Even though OLS gives significant results, we do not claim anticipation effects considering the selection issue.
↵b Ag denotes agricultural land cover, except hay.
↵* p < 0.1;
↵** p < 0.05;
*** p < 0.01.

View popup

Table 5

Estimation of the Total Impacts of Nonresident Owners

Applied Counties: Top 4% (27 Counties)^{^a}		2002	2012	Change (2002–2012)
Total acreage underutilized if	15% of the agricultural parcels owned by nonresidents, 5% to be owned by nonresidents (emerging second-home market; i.e., Columbia)	34,045	45,393	11,348
Total acreage underutilized if	65% of the agricultural parcels owned by nonresidents (saturated second-home market; i.e., Ulster)	147,527	<147,527	<0
Total acreage (USDA National Agricultural Statistics Service)	Land in production (harvesting field crops/hay/vegetables)	986,639	916,428	−70,211
	Total land in farms	2,269,543	2,236,128	−33,415

Note: Outcomes are lower bound estimations of total underutilized acreage land in farms in applicable counties if farmland is owned by nonresidents instead of farmers. The applicable counties are selected based on the ranking list in Appendix F. The decrease of total land in farms is closely related to farmland conversion, which is usually used to explain the decrease in agricultural production. The decrease of land in production incorporates farmland conversion and farmland underutilization.
↵a The baseline is the 2002 data from USDA NASS, at https://www.nass.usda.gov/.