Upload
gabriella-stewart
View
212
Download
0
Embed Size (px)
Citation preview
1
20th SIEP Conference University of Pavia
25-26 September 2008
Evaluating the Efficiency of Italian Penitentiaries
Fabrizio Balassone, Italian Ministry of Economy and Banca d’ItaliaMarco Camilletti, Italian Ministry of EconomyVeronica Grembi, Catholic University of Milan
Alberto Zanardi, University of Bologna
2
The Italian penitentiary system: main issues
Italian prisons are over crowded. Over 1995-2005 :
– stark increase of inmates population (+22,2%)
– prisons capacity almost stable
How to tackle this issue?
– 2006 pardon, but again overcrowding by 2008 (pop08=92.5% pop05)
– alternative detention measures
– increasing capacity by building more prisons
Strict budget constraint → need to enhance efficiency
But there is scant evidence on the economic performance of Italian prisons
3
Aim and structure of the paper
This paper aims to empirically assess the technical efficiency of Italian penitentiaries
- The Italian penitentiary system: main facts
- Literature review
- Our analysis:
* data description
* econometric model
* estimation results
- Conclusions: policy implications and future research
4
The Italian penitentiary system: main facts (1)
The majority of facilities are located in the South
T erritorial distribution of penitentiary facilities by type (year 2005)
92 75 70
180156
39
39
0
50
100
150
200
250
North-West North-E ast C entre S outh andIslands
District Houses S entence S erving Institutes
5
The Italian penitentiary system: main facts (2)
- Average capacity higher in the North- Overall capacity higher in the South
C ertified normal accomodation by Macroregion(year 2005)
0
50
100
150
200
250
300
North-West North-E ast C entre S outh andIs lands
0
3000
6000
9000
12000
15000
18000
Average Total
6
The Italian penitentiary system: main facts (3)
Overcrowding more marked in the North
Overcrowding (average )
Index of overcrowding<0.5
(% of penitentiaries )
Index of overcrowding>1.2
(% of penitentiaries )North-West 1.5 1.9 78.5North-East 1.5 0.0 80.2Centre 1.2 4.6 53.2South and Islands 1.2 5.0 55.3Source: Ministry of justice
Penitentiary facilities distribution by Index of overcrowding and Macroregion (year 2005 )
7
The Italian penitentiary system: main facts (4)
Prisons in the South tend to be over-staffed
Police/Inmate by Macroregion (year 2005 )
0.0
0.2
0.4
0.6
0.8
1.0
1.2
North-West North-E ast C entre S outh andIs lands
8
The Italian penitentiary system: main facts (5)
- Average cost of custody widely dispersed
- Sharply declining with prison size: economies of scale or overcrowding?0
500
001
0000
01
5000
02
0000
0
Ave
rag
e ex
pen
ditu
re p
er
inm
ate
0 500 1000 1500
Inmates
9
Literature review (1)
Limited work on empirical analysis of prison efficiency
(A) Non-parametric approach (DEA)
UK: Ganley and Cubbin (1992)- 33 prisons, 1 year- control for overcrowding (and other variables)- technical inefficiency accounts for 12% of expenditure
US: Butler and Johnson (1997)- 22 prisons, 1 year- do not control for overcrowding- inefficiency accounts for up to 50% of expenditure
10
Literature review (2)
(B) Parametric approach (cost and/or production function estimation)
US: Trumbull and Witte (1981) - very limited sample- no frontier analysis- no prices- unexploited economies of scale
IT: Panci (1999)- 189 prisons, 1 year- no prices- no control for overcrowding- 14% average inefficiency (higher in the South)
11
Data
Source: Italian Department of Prisons Administration
Information by prison over 2003-2007 on:- outputs (number and attributes of inmates),- inputs (workforce and other expenses),- prices (salaries),- structural characteristics (capacity, opening date, conditions, etc.)
2006 legal pardon → structural break of the data, limitation of dataset to 2003-2005
Outliers → 4.6% obs. dropped → 435 annual obs.
12
The econometric model
- Short-term cost function (no capital input)
- Log-Log specification (all variables normalized by their own average)
- Several controls tried (on quality of inputs & outputs): none significant
- No quadratic term
- Stochastic frontier estimation for panel data (ML estimation)
tc total costs (wages + other expenses)ti total number of inmatesaw average wage of police unitsoc overcrowding index (number of inmates/normal capacity) v iid disturbance termu iid time-invariant truncated non-negative disturbance term
→ technical efficiency index
uoc
ocaw
awti
titc
tciit
i tit
it
i tit
it
i tit
it
i tit
it lnlnlnln
3210
13
Estimation results (1)
- All explanatory variables significant and signed as expected
- Large average technical inefficiency (much larger without controlling for liia)
ln ti^ 0.81*** (29.13)
ln aw^ 1.01*** (18.17)
ln oc^ -0.77*** (-24.66)
constant -0.87*** (-8.49)
Average inefficiency O bservations Number of G roups L og L ikelihood s igma u s igma v C hisq 1244
z s tatistics in parentheses *** p<0.001, ** p<0.01, * p<0.05
2.48
435
157
397.6
0.267
0.0415
x ̂= x/average(x)
14
Estimation results (2)
Unexploited economies of scale - controlling for overcrowding
- even using inefficient technology
Averag e expenditure per inmate (2005)
05000
1000015000200002500030000350004000045000
25 150 275 400 525 650 775 900 1025 1150
Inmates population
lower bound central upper bound OL S
15
Estimation results (3)
- technical inefficiency does not vary much across regions
- technical inefficiency higher than average is related to: a) size b) police/inmates ratio c) overcrowding
Area Avg. Inefficiency North West 2,39 North East 2,57 Center 2,40 South and Islands 2,53
Low ineff. High ineff.
Capacity 199.2 216.1
Inmates 244.2 309.5
Overcrowding 1.3 1.4
Average wage 27107.8 26853.1
Police/Inmate 0.7 1.2
16
Concluding remarks
Policy implications
• Most prisons are undersized w.r.t. optimal scale → long-term programme to increase the average capacity of prisons
• Large technical inefficiency of most prisons → possible short-term measures to overcome the most striking cases of over-staffing
Future research
• Estimation of possible expenditure reductions stemming from closing small-size prisons and improving efficiency
• Robustness of results if more adequate proxies of output and input are used
• Check results against non-parametric estimates (DEA)
• Definition of output (functions other than detention)