Appendix C: Equitable Sharing Methods

Analysis of equitable sharing requests

For our analysis of equitable sharing requests, we evaluated (1) how reliably the U.S. Department of Justice grants state and local agencies’ requests for a share of equitable sharing proceeds, (2) how reliably the DOJ approves the specific percentage of proceeds state and local agencies request, and (3) what share of proceeds state and local agencies receive overall.

The unit of analysis is an equitable sharing request submitted by a state or local law enforcement agency within the 50 states and the District of Columbia. Federal, tribal, and foreign agencies are excluded from the analyses. The sample comprises all requests with a decision rendered between 2000 and 2023. All data are from the Department of Justice’s Consolidated Asset Tracking System.

How reliably does the DOJ grant equitable sharing requests?

To determine how reliably the DOJ grants state and local agencies’ requests for equitable sharing proceeds, we computed the total number of requests in our sample and disaggregated it by type of decision rendered. Table C1 presents requests by decision type across the entire data period. Granted requests are those for which any sharing of proceeds (net expenses) was approved.

Table C1: Number and percentage of decisions on equitable sharing requests, by type, 2000–2023

Decision on request 1	Number	Percent
Granted	652,903	92.3%
Rejected – No property/proceeds	46,507	6.6%
Rejected – No memo	2,535	0.4%
Rejected – State law prohibits	2,156	0.3%
Rejected – Failed to confirm	903	0.1%
Other – Extinguished by AFMLS*	2,731	0.4%
Total requests, 2000–2023	707,735

*Asset Forfeiture and Money Laundering Section

How reliably does the DOJ approve the requested percentage?

To determine how reliably the DOJ approves the specific percentage of proceeds state and local agencies request, we used two comparable CATS variables. The first, the requested percentage (shr_req_pct), measures the percentage of an asset’s value that a participating state or local agency requested. The other, the approved percentage (shr_pct_to_shr), measures the percentage of the asset’s value that the relevant DOJ agency approved sharing with the agency. ²  

The requested percentage variable originally comes from the Equitable Sharing Request Form, or DAG-71, which agencies use to request sharing on a particular asset—specifically an old version of the form that includes a field for the specific percentage of net proceeds an agency is requesting. The “new” DAG-71, introduced August 2014, has no such field. Agencies could and did use either form across all years in our analysis. Table C2 shows the number of equitable sharing requests submitted by year, disaggregated by which form was used.

Table C2: Equitable sharing requests by form type, 2000–2023

Year	Equitable sharing request form
	Old	New	Total
2000	26,116	N/A	26,116
2001	26,927	N/A	26,927
2002	21,141	N/A	21,141
2003	30,999	N/A	30,999
2004	29,000	N/A	29,000
2005	28,124	N/A	28,124
2006	31,959	N/A	31,959
2007	33,717	N/A	33,717
2008	32,316	N/A	32,316
2009	34,941	N/A	34,941
2010	35,124	N/A	35,124
2011	37,358	N/A	37,358
2012	34,624	N/A	34,624
2013	47,842	N/A	47,842
2014	39,586	1,824	41,410
2015	19,541	22,997	42,538
2016	6,665	28,066	34,731
2017	2,293	17,703	19,996
2018	2,470	29,755	32,225
2019	795	24,379	25,174
2020	713	9,879	10,592
2021	407	11,261	11,668
2022	236	23,771	24,007
2023	209	14,997	15,206
Total	523,103	184,632	707,735

Because the requested percentage appears only if an agency used the old form, we restricted our analysis to the 73.9% of requests submitted using that form. Additionally, we excluded the 5.9% of requests where (1) the percentage requested was zero or (2) there were no remaining proceeds to share after other agencies (federal, state, and local) took their share. In this way, we ensured we were examining only requests for a positive share of an asset’s proceeds and that the assets associated with requests had available proceeds to share after expenses and other claims were paid.

We subtracted the requested percentage from the approved percentage to arrive at the “gap” between the two. We grouped the subsample by whether the gap was positive, zero, or negative to observe the frequency of each type of gap. And because the requested percentage and the approved percentage vary based on the number of agencies requesting shares of a single asset, we also grouped the subsample by whether requests were made by a single agency, two to three agencies, or four or more agencies. For each group, we computed the median approved percentage, the median requested percentage, and the median gap in percentage points (see Table C3).

Table C3: Gaps between requested percentage and the approved percentage, by number of agencies requesting shares of an asset and whether the gap was positive, zero, or negative

Positive gap (DOJ approved a larger share than requested)
Agencies	No. requests with positive gap	% requests with positive gap	Median approved percentage	Median requested percentage	Median gap in percentage points
1 agency	1,632	7.6%	80.0	40.0	40.0
2–3 agencies	5,252	24.5%	35.0	20.0	10.0
4+ agencies	14,538	67.9%	11.7	9.0	2.0
Total with positive gap	21,422	100.0%	35.0	20.0	10.0

Zero gap (DOJ approved the exact share requested)
Agencies	No. requests with zero gap	% requests with zero gap	Median approved percentage	Median requested percentage	Median gap in percentage points
1 agency	60,266	15.7%	80.0	80.0	0.0
2–3 agencies	83,642	21.8%	30.0	30.0	0.0
4+ agencies	240,600	62.6%	8.0	8.0	0.0
Total with zero gap	384,508	100.0%	30.0	30.0	0.0

Negative gap (DOJ approved a smaller share than requested)
Agencies	No. requests with negative gap	% requests with negative gap	Median approved percentage	Median requested percentage	Median gap in percentage points
1 agency	8,023	9.3%	80.0	100.0	-20.0
2–3 agencies	23,388	27.2%	30.0	40.0	-10.0
4+ agencies	54,712	63.5%	8.6	16.0	-5.0
Total with negative gap	86,123	100.0%	30.0	40.0	-10.0

It should be noted that the percentage of an asset’s proceeds that an agency actually receives may be lower than the approved percentage if, for instance, there were expenses associated with forfeiting the asset or proceeds were reduced or exhausted after sharing with other agencies. To examine how closely the percentage an agency actually received tracks the requested percentage and the approved percentage, we calculated the actual percentage received by dividing the amount of proceeds received by an agency by the total amount of proceeds available for sharing from that asset after expenses and other claims were paid. ³  

For this analysis, we confined our data to granted requests made by agencies that submitted at least one form with a requested percentage both before and after 2014, when the new DAG-71 form was introduced. In this way, we excluded agencies that fell out of the subsample after the new form was introduced, were not present in the subsample before the new form was introduced, or fully adopted the new form after it was introduced. In doing this, we tried to narrow the variability in unobservable agency-specific characteristics that affect both their equitable sharing requests and their adoption of the new form. We ended with a comparable subset of agencies for the whole period, representing 49% of the state and local law enforcement agencies in the overall sample of all requests. For all requests in this subset, we computed annual averages across agencies of the requested percentage, the approved percentage, and the actual percentage received. The results are grouped by how many agencies submitted requests for an asset (see Figure C1). All three metrics track closely for most of the time series, particularly for assets with requests from four or more agencies.

Figure C1: Requested percentage, approved percentage, and actual percentage received by number of agencies requesting shares of an asset, 2000–2023

We also explored whether there was a relationship between the percentage of an asset’s value that the relevant DOJ agency approved sharing with an agency and that agency’s contributions toward seizing or forfeiting that asset. For each submitted request in our overall sample of all requests, we computed relative measures of both hours and expenses incurred by the requesting agency that were directly related to processing or litigating the seized asset. Specifically, we calculated the number of hours contributed by the requesting agency as a percentage of the sum of hours contributed by all local agencies requesting a share of proceeds for a given asset. Likewise, we computed the expenses incurred by the requesting agency as a percentage of the sum of expenses incurred by all local agencies requesting a share. We then estimated the correlation between our relative measures of contributed effort and the approved percentage, distinguishing between adoptions and joint investigations (see Table C4). All correlations were strong, suggesting agencies can expect the relevant DOJ agency to approve shares roughly proportional to their reported contributed hours and expenses.

Table C4: Correlations between the approved percentage and the percentage of contributed hours and expenses, by sharing type

	Correlation coefficient		Sharing type	No. requests by sharing type
	% contributed hours	% contributed expenses
Approved percentage	0.891	0.850	Adoption	83,538
	0.787	0.832	Joint	624,077
	0.832	0.869	All	707,615

What share of proceeds do state and local agencies receive overall?

To determine what share of equitable sharing proceeds state and local agencies receive overall, we computed, for each request granted in a given year, the amount received by an agency as a percentage of the total proceeds available for sharing after expenses and other claims were paid. This is the same calculation used to calculate the actual percentage received that we report above, but for the overall sample of all requests. We also computed the average percentage received across all agencies. Figure C2 shows the annual average percentage grouped by how many agencies submitted requests for a share of asset proceeds.

Figure C2: Percentage of asset proceeds received per agency, by number of agencies requesting shares of an asset

Analysis of efforts to reform equitable sharing

We evaluated whether the 2015 federal reform banning most adoptive forfeitures affected equitable sharing nationwide and whether anticircumvention reforms affected equitable sharing in the states that enacted them. We used segmented regression techniques to compare the periods before and after a reform and assessed whether the policy interventions resulted in a break in the trend of the relevant outcome. All models use cluster-robust standard errors estimated with the xtitsa package in Stata, which uses a generalized estimating equations framework to account for within-unit correlation over time.

The unit of analysis is a state-month. Our sample consists of monthly data from January 2000 to December 2021 for all states and the District of Columbia, except for Maine. ⁴  Equitable sharing data came from CATS. Covariate data came from the Bureau of Labor Statistics, the Census Bureau, or the Federal Bureau of Investigation’s Uniform Crime Reporting Program (see Table C5).

All outcome variables were computed based on assets shared with a state or local law enforcement agency, regardless of the number of requests for an asset. An asset was assigned to a state in which at least one state or local law enforcement agency had a request granted for it. An asset with multiple requests from a given state was counted only once in that state. An asset with multiple granted requests from various states was counted only once in each state. ⁵  Each asset was assigned to the month and year when it was seized, not when proceeds were paid, ensuring the count of shared assets reflects the federal and state statutes in place at the time of seizure.

Table C5: Variables and data sources

Variable	Measure	Source
Number of shared assets	Number of federally forfeited shared assets	CATS
Percent assets below threshold	Percentage of assets below a given threshold value	CATS
Unemployment	Seasonally adjusted monthly state unemployment rate	BLS
Population	Monthly state population, linearly interpolated using annual end-of-year population estimates	Census Bureau
Crime	Monthly number of total crimes (violent plus property crime) per 100k people	FBI UCR Program
Police	Monthly number of sworn law enforcement officers per 100k people	FBI UCR Program Law Enforcement Employees data

Analysis of federal reform

We used Interrupted Time Series analysis to evaluate the effect of the DOJ 2015 reform on equitable sharing nationwide. We estimated the following model: ⁶

Y_it = β₀ + β₁Time_t + β₂Post_t + β₃Post_t ∙Time_t + β₄W_it + β₅D_i + β₆D_i∙Time_t + β₇C_t + ε_it,

where Y is the outcome variable, i is a state, t is the month, Time is a time trend, Post is a time dummy variable equal to 1 after January 2015 and 0 before, W is a vector of covariates identified in Table C5, and ε is an error term. Further, we included a vector, D, of 10 dummy variables (one for each state with an anticircumvention reform) and an interaction term, D∙Time, capturing each state’s post-reform trend. Finally, when the outcome variable is the number of shared assets, we included a dummy variable, C, equal to 1 from April to December 2020 to control for the disruption due to the COVID-19 pandemic. ⁷  

Depending on the outcome variable, we estimated two versions of the ITS model: In Model 1, the outcome variable was the number of shared assets; in Model 2, we defined the outcome variable as the percentage of shared assets with a value below $50,000 at the time of seizure. ⁸  

Table C6 presents basic descriptive statistics for the number of shared assets and control variables, with the observations covering all states across all years, with the exception of Maine. The results of the ITS estimations for the two models are presented in Table C7. ⁹

Table C6: Summary statistics, DOJ 2015 reform analysis, 2000–2021

Variable	Observations	Mean	Std. deviation
Number of shared assets	13,200	16.5	25.3
Percent assets below $50k	11,886	81.8	20.0
Unemployment rate	13,200	5.6	2.2
Population	13,200	6,129,765	6,822,461
Crime rate per 100k people	13,200	3,318	1,034
Sworn officers per 100k people	13,200	221	82

Table C7: Coefficient estimates for the DOJ reform ITS models

			Model 1		Model 2
			No. shared assets		% shared assets under $50k
	Coef.	Variable	Coef.	P>|z|	Coef.	P>|z|
	β0	Constant	-90.6672	0.3100	127.8335	0.0000
	β1	Pre –reform trend	0.0469	0.0040	-0.0227	0.0000
	β2	Discrete effect	-8.1716	0.0000	-3.8075	0.0000
	β3	Reform trend effect	-0.0650	0.0110	0.0137	0.3510
	β4	Unemployment rate	0.2658	0.1960	-0.3485	0.0010
		Log population	6.6443	0.2720	-2.0958	0.0100
		Crime rate per 100k	0.0027	0.0180	-0.0018	0.0010
		Log sworn officers per 100k	-1.0934	0.4790	-0.3627	0.7580
AZ	β5	Discrete state effect	-4.6194	0.0040	-1.3969	0.7970
	β6	State trend effect	0.0364	0.1250	0.1207	0.4860
CA	β5	Discrete state effect	-59.4328	0.0000	-8.7310	0.0890
	β6	State trend effect	-0.7776	0.0000	-0.0808	0.5750
CO	β5	Discrete state effect	-5.0766	0.0000	-22.5285	0.0000
	β6	State trend effect	-0.1478	0.0000	-0.2805	0.1100
DC	β5	Discrete state effect	1.7196	0.2220
	β6	State trend effect	0.0403	0.0630
MD	β5	Discrete state effect	-25.5328	0.0000	-9.9143	0.0520
	β6	State trend effect	-0.0721	0.0010	-0.0015	0.9920
NE	β5	Discrete state effect	-2.1209	0.0510	-12.2920	0.0120
	β6	State trend effect	0.0107	0.5390	0.2048	0.1000
NM	β5	Discrete state effect	-8.3491	0.0000
	β6	State trend effect	0.0470	0.0610
OH	β5	Discrete state effect	0.5877	0.6390	-0.8256	0.8750
	β6	State trend effect	0.0441	0.0410	0.0877	0.5730
PA	β5	Discrete state effect	5.8577	0.0000	-2.2892	0.6710
	β6	State trend effect	-0.2951	0.0000	-0.1470	0.3850
WI	β5	Discrete state effect	-13.3111	0.0000	0.2408	0.9670
	β6	State trend effect	0.1571	0.0000	0.0608	0.7850
	β7	COVID-19 dummy	0.6445	0.4560

Analysis of state reforms

We used Comparative Interrupted Time Series analysis to evaluate the effect of state anticircumvention reforms on equitable sharing, again using generalized estimating equations and robust standard errors clustered at the state level. Specifically, we conducted separate CITS analyses for eight reform states: Arizona, California, Colorado, Maryland, Nebraska, Ohio, Pennsylvania, and Wisconsin. ¹⁰   In each case, we took one reform state as the treatment group and the set of 40 states that have never implemented an anticircumvention reform as our control group. ¹¹  

For a given reform state, the CITS model estimated takes the form:

Y_it = β₀ + β₁Time_t + β₂Post_t + β₃Post_t ∙Time_t + β₄Treat_i + β₅Treat_i ∙Time_t + β₆Treat_i ∙Post_t + β₇Treat_i  ∙Post_t  ∙Time_t  + β₈W_it + ε_it,

where Y is the outcome variable, i is a state, t is the month, Time is a time trend, Post is a time dummy variable equal to 1 after the anticircumvention reform’s effective date and 0 before, Treat is a treatment dummy variable equal to 1 if i is the reform state and 0 otherwise, W is the vector of covariates discussed earlier, and ε is an error term. Post∙Time, Treat∙Time, Treat∙Post, Treat∙Post∙Time, and D∙Time are interaction terms. The dummy variable Post is defined using the state reform’s effective date, listed in Table C8.

Table C8: Effective dates of state anticircumvention reforms

State	Effective date
Arizona	July 2017
California	January 2017
Colorado	August 2017
Maryland	October 2016
Nebraska	July 2016
Ohio	April 2017
Pennsylvania	July 2017
Wisconsin	April 2018

Depending on the outcome variable, we estimated two versions of our CITS model. In Model 3, the outcome variable is the number of shared assets. In this case, we added to the right-hand side both a time dummy variable, D_t, equal to 1 after January 2015 and 0 before to control for the DOJ reform, and an interaction term, D∙Time. We estimated Model 3 for all eight reform states.

In Model 4, the outcome variable is the percentage of shared assets below the relevant state threshold. We estimated this model for the six states whose anticircumvention reforms established a minimum threshold for transferring funds to the federal government or receiving equitable sharing funds; we excluded Pennsylvania and Wisconsin as their reforms set no such thresholds. Table C9 summarizes the relevant value variable, the applicable asset type, and the threshold level for the six states under consideration.

Table C10 presents summary statistics for the outcome variable(s) in each state. The results of the CITS estimations for Models 3 and 4 are presented in Table C11.

Table C9: Characteristics of anticircumvention threshold reforms

State	Legal restriction on	Value variable	Asset type	Threshold
Arizona	Transferring assets	Asset value at seizure	All	$75,000
California	Receiving funds	Forfeiture amount	Cash and financial instruments	$40,000
Colorado	Receiving funds	Forfeiture amount	All	$50,000
Maryland	Transferring assets	Asset value at seizure	Cash	$50,000
Nebraska	Transferring assets	Asset value at seizure	All	$25,000
Ohio	Transferring assets	Asset value at seizure	All	$100,000

Table C10: Summary statistics for the outcome variable, 2000–2021

Reform state	No. control states	Variable	Number of state-months 12	Mean	Std. deviation
Arizona	40	No. shared assets	10,824	13.1	17.5
		% shared assets under $75k	9,687	85.7	18.4
California	40	No. shared assets	10,824	15.9	26.8
		% shared assets under $40k	9,688	65.4	25.6
Colorado	40	No. shared assets	10,824	13.1	17.5
		% shared assets under $50k	9,686	81.1	21.0
Maryland	40	No. shared assets	10,824	13.5	17.9
		% shared assets under $50k	9,680	64.6	26.1
Nebraska	40	No. shared assets	10,824	13.0	17.4
		% shared assets under $25k	9,688	67.5	25.4
Ohio	40	No. shared assets	10,824	13.5	17.8
		% shared assets under $100k	9,688	88.9	16.2
Pennsylvania	40	No. shared assets	10,824	13.4	17.6
Wisconsin	40	No. shared assets	10,824	13.5	17.7

Table C11: Coefficient estimates of the CITS models for reform states

Arizona		Model 3		Model 4
		No. shared assets		% shared assets under $75k
Coef.	Variable	Coef.	P>|z|	Coef.	P>|z|
β0	Constant	-11.2554	0.0000	119.6948	0.0000
β1	Pre-period trend	0.0004	0.6250	-0.0148	0.0700
β2	Level diff at baseline	0.3150	0.1560	0.3932	0.8650
β3	Trend diff at baseline	-0.0026	0.0160	0.0260	0.0260
β4	Level change in control	0.2690	0.0000	4.3677	0.0220
β5	Trend change in control	-0.0038	0.3350	0.0482	0.5510
β6	Level diff in treatment	-0.9256	0.0000	-12.0574	0.0000
β7	Trend diff in treatment	0.0153	0.0000	0.3146	0.0000
β8	Unemployment rate	0.0082	0.5340	-0.2756	0.0590
	Log population	0.9442	0.0000	-1.9535	0.0080
	Crime rate	-0.0001	0.5110	-0.0011	0.2080
	Log sworn officers	-0.0964	0.6720	0.5852	0.7740
	DOJ reform dummy	-0.3907	0.0000	-0.9802	0.4720
	DOJ reform trend	-0.0108	0.0020	-0.0949	0.2430

California		Model 3		Model 4
		No. shared assets		% shared cash assets under $40k*
Coef.	Variable	Coef.	P>|z|	Coef.	P>|z|
β0	Constant	-11.2446	0.0000	110.9381	0.0000
β1	Pre-period trend	0.0004	0.5930	-0.0198	0.1870
β2	Level diff at baseline	0.1929	0.3340	14.2224	0.0000
β3	Trend diff at baseline	0.0028	0.0000	-0.0117	0.2040
β4	Level change in control	0.3224	0.0000	2.8432	0.2760
β5	Trend change in control	0.0066	0.1430	0.0957	0.4580
β6	Level diff in treatment	-0.4910	0.0000	-13.4447	0.0000
β7	Trend diff in treatment	-0.0060	0.0020	-0.0522	0.2880
β8	Unemployment rate	0.0075	0.5610	-0.6543	0.0000
	Log population	0.9445	0.0000	-2.4131	0.0310
	Crime rate	-0.0001	0.5150	-0.0030	0.0340
	Log sworn officers	-0.0992	0.6650	1.1721	0.7070
	DOJ reform dummy	-0.3099	0.0000	-0.4719	0.8470
	DOJ reform trend	-0.0190	0.0000	-0.0325	0.8170

* Including financial instruments.

Colorado		Model 3		Model 4
		No. shared assets		% shared assets under $50k
Coef.	Variable	Coef.	P>|z|	Coef.	P>|z|
β0	Constant	-11.2615	0.0000	128.7571	0.0000
β1	Pre-period trend	0.0004	0.6080	-0.0282	0.0030
β2	Level diff at baseline	-0.3589	0.0000	5.9423	0.0000
β3	Trend diff at baseline	0.0060	0.0000	0.0253	0.0010
β4	Level change in control	0.2460	0.0000	5.4440	0.0050
β5	Trend change in control	-0.0056	0.1410	0.0143	0.8590
β6	Level diff in treatment	-1.3262	0.0000	-29.4841	0.0000
β7	Trend diff in treatment	-0.0071	0.0300	-0.1945	0.0030
β8	Unemployment rate	0.0078	0.5560	-0.2643	0.0530
	Log population	0.9444	0.0000	-2.3305	0.0020
	Crime rate	-0.0001	0.5110	-0.0019	0.0650
	Log sworn officers	-0.0956	0.6750	-0.0410	0.9860
	DOJ reform dummy	-0.4162	0.0000	-2.5551	0.0930
	DOJ reform trend	-0.0092	0.0030	-0.0845	0.2740

Maryland		Model 3		Model 4
		No. shared assets		% shared cash assets under $50k
Coef.	Variable	Coef.	P>|z|	Coef.	P>|z|
β0	Constant	-11.2691	0.0000	121.1414	0.0000
β1	Pre-period trend	0.0005	0.5450	-0.0247	0.1080
β2	Level diff at baseline	1.3839	0.0000	20.8838	0.0000
β3	Trend diff at baseline	-0.0040	0.0000	-0.0631	0.0000
β4	Level change in control	0.2278	0.0000	2.1212	0.3270
β5	Trend change in control	0.0062	0.2320	0.1500	0.2980
β6	Level diff in treatment	-0.7927	0.0000	-5.7164	0.0000
β7	Trend diff in treatment	-0.0125	0.0000	-0.0052	0.9150
β8	Unemployment rate	0.0067	0.6170	-0.4344	0.0080
	Log population	0.9443	0.0000	-2.9307	0.0080
	Crime rate	-0.0001	0.5180	-0.0021	0.1480
	Log sworn officers	-0.0941	0.6790	-0.3193	0.9090
	DOJ reform dummy	-0.3411	0.0000	1.0866	0.6900
	DOJ reform trend	-0.0172	0.0000	-0.0083	0.9580

Nebraska		Model 3		Model 4
		No. shared assets		% shared assets under $25k
Coef.	Variable	Coef.	P>|z|	Coef.	P>|z|
β0	Constant	-11.2876	0.0000	123.3546	0.0000
β1	Pre-period trend	0.0005	0.5520	-0.0296	0.0120
β2	Level diff at baseline	1.0996	0.0000	6.7481	0.0000
β3	Trend diff at baseline	-0.0018	0.0020	0.0062	0.5730
β4	Level change in control	0.0365	0.6130	1.7225	0.4950
β5	Trend change in control	-0.0006	0.9330	0.2997	0.2340
β6	Level diff in treatment	-0.3702	0.0000	-24.5040	0.0000
β7	Trend diff in treatment	0.0052	0.0010	0.3353	0.0000
β8	Unemployment rate	0.0060	0.6550	-0.4685	0.0050
	Log population	0.9443	0.0000	-3.1327	0.0040
	Crime rate	-0.0001	0.5200	-0.0025	0.0670
	Log sworn officers	-0.0901	0.6910	1.5122	0.5870
	DOJ reform dummy	-0.3874	0.0000	-2.7346	0.3190
	DOJ reform trend	-0.0089	0.1570	-0.3118	0.2090

Ohio		Model 3		Model 4
		No. shared assets		% shared assets under $100k
Coef.	Variable	Coef.	P>|z|	Coef.	P>|z|
β0	Constant	-11.2633	0.0000	116.7088	0.0000
β1	Pre-period trend	0.0004	0.6290	-0.0152	0.0340
β2	Level diff at baseline	-0.0884	0.4990	3.1624	0.0080
β3	Trend diff at baseline	0.0018	0.0530	0.0255	0.0010
β4	Level change in control	0.2784	0.0000	2.5045	0.1140
β5	Trend change in control	0.0000	0.9980	0.0069	0.9370
β6	Level diff in treatment	-0.1250	0.0360	-4.0484	0.0000
β7	Trend diff in treatment	0.0102	0.0000	0.0503	0.1510
β8	Unemployment rate	0.0086	0.5160	-0.1864	0.0770
	Log population	0.9438	0.0000	-1.6071	0.0060
	Crime rate	-0.0001	0.5130	-0.0011	0.1330
	Log sworn officers	-0.0942	0.6760	0.5905	0.7310
	DOJ reform dummy	-0.3597	0.0000	-0.8833	0.4970
	DOJ reform trend	-0.0134	0.0000	-0.0271	0.7470

Pennsylvania		Model 3
		No. shared assets
Coef.	Variable	Coef.	P>|z|
β0	Constant	-11.2617	0.0000
β1	Pre-period trend	0.0003	0.6510
β2	Level diff at baseline	-0.2092	0.1830
β3	Trend diff at baseline	0.0006	0.2220
β4	Level change in control	0.2631	0.0000
β5	Trend change in control	-0.0038	0.3360
β6	Level diff in treatment	0.2140	0.0000
β7	Trend diff in treatment	-0.0025	0.4160
β8	Unemployment rate	0.0079	0.5500
	Log population	0.9441	0.0000
	Crime rate	-0.0001	0.5130
	Log sworn officers	-0.0943	0.6780
	DOJ reform dummy	-0.3804	0.0000
	DOJ reform trend	-0.0108	0.0020

Wisconsin		Model 3
		No. shared assets
Coef.	Variable	Coef.	P>|z|
β0	Constant	-11.2655	0.0000
β1	Pre-period trend	0.0004	0.6350
β2	Level diff at baseline	1.2633	0.0000
β3	Trend diff at baseline	-0.0056	0.0000
β4	Level change in control	0.0258	0.7030
β5	Trend change in control	-0.0124	0.0000
β6	Level diff in treatment	0.0033	0.9580
β7	Trend diff in treatment	0.0325	0.0000
β8	Unemployment rate	0.0094	0.4790
	Log population	0.9437	0.0000
	Crime rate	-0.0001	0.5100
	Log sworn officers	-0.0939	0.6790
	DOJ reform dummy	-0.4907	0.0000
	DOJ reform trend	-0.0028	0.2380