Download - The performance of the public sector Pierre Pestieau CREPP, University of Liège, CORE, PSE and CEPR
The performance
of the public sector
Pierre PestieauCREPP, University of Liège,
CORE, PSE and CEPR
2
Outline
1. Introduction
2.The performance approach and the concept of best practice
3. Measuring productive efficiency
4. The performance of social protection
5. Conclusion
3
1. Introduction
Measuring and ranking: a must
People do it anyway but badly Transparency and governance Yardstick competition – Open Method of Coordination (OMC)
Important distinction between the public sector as a whole and its components
Problem of aggregation Technical link between outcomes (outputs) and resources (inputs)
The performance is to be measured by the extent to which the preassigned objectives are achieved.
4
2. The performance approach and the concept of best practice
The public sector is a set of more or less aggregated production units (social security administration, railways, health care, education, national defence, social protection,…)
Each unit is supposed to use a number of resources, within a particular setting, to produce a number of outputs
Those outputs are related to the objectives that have been assigned to the production unit by the principal, the authority in charge
Approach used here: productive efficiency and to measure it, the efficiency frontier technique is going to be used
5
Productive efficiency is just a part of an overall performance analysis. It has two advantages:
It can be measured It is a necessary condition for any other type of
objectives
Main drawback: it is relative
Based on a comparison among a number of rather similar production units
Its quality depends on the quality of the observation units.
6
Set of observations Best practice frontier
Non parametric method: DEA (data envelopment analysis)
Parametric method
Comparative advantage
Illustration with one input/one output
7
output
input
Set of comparable observations
Figure 1
8
output
input
Parametric
Figure 2
9
output
input
Non Parametric
Figure 3
10
output
input
bc
A B
a
t+1
t
Figure 4
11
Technical progress:
Efficiency in t:
in t + 1:
aA
A
bB
B
Change in efficiency: ca -
12
Motivation of efficiency study: performance improvement
Factors of inefficiency:
Exogenous (location) Endogenous (low effort) Policy related (ownership, competition)
13
3. Measuring productive efficiency. Conceptual and data problems
Two problems.
Weak link between the inputs used and the expected outcomes
Confusion between lack of data and conceptual difficulties
Research strategy. Two areas quite typical of public spending: education and railways transports; how performance should be measured if data availability were not a constraint?More precisely, when listing the outputs and the inputs, assume that the best evidence one can dream of is available.
14
3.1. The best evidence
Inter-country comparison.
Importance of institutional, political and
geographical factors.
15
Ideal data Outputs Passenger kilometres Comfort and punctuality Freight tons and kilometers - bulk - containers - others Delivery quality and punctuality Equity of access Passengers per seat Inputs Labor (disaggregated) Equipment (disaggregated by type and by
quality) Tracks (length and quality) Energy (sources) Environment Geography, stage length Autonomy Competition or contestability Price discrimination Community service obligation Observations Very large number of years and countries
Railways
16
Ideal data
Output Acquired skills (of sample of 18y. old individuals)
- math, science, reading - foreign languages Direct employability Indirect employability (through college) Happiness Contribution to R and D
Input Teachers (level and quality) Staff Building, equipment Spatial distribution of schools Skills at the end of the primary education level
Environment Competition between networks Competition with private schools Role of the family Unemployment rate, economic growth Pedagogical technique
Observations Large number of countries and years
High schools
17
3.2. Actual studies
Most qualitative variables are missing.
Difference between developed and less developed countries.
Focus on financial variables.
18
Ideal data Data used
in recent studies Outputs Passenger kilometres v Comfort and punctuality ~ Freight tons and kilometers v - bulk ~ - containers ~ - others ~ Delivery quality and punctuality ~ Equity of access – Passengers per seat ~ Inputs Labor (disaggregated) v Equipment (disaggregated by type and by
quality) v
Tracks (length and quality) ~ Energy (sources) ~ Environment Geography, stage length ~ Autonomy ~ Competition or contestability ~ Price discrimination ~ Community service obligation ~ Observations Very large number of years and countries Too small
Note: v = OK; ~ = more or less; – = unavailable
Railways
19
Ideal data Recent studies
Output Acquired skills (of sample of 18y. old individuals)
- math, science, reading v - foreign languages – Direct employability – Indirect employability (through college) ~ Happiness – Contribution to R and D ~
Input Teachers (level and quality) ~ Staff ~ Building, equipment v Spatial distribution of schools – Skills at the end of the primary education level –
Environment Competition between networks Competition with private schools ~ Role of the family – Unemployment rate, economic growth ~ Pedagogical technique ~
Observations Large number of countries and years
~
Note: v = OK; ~ = more or less; – = unavailable
Education
20
Sector and authors Number of units
Type and period of data
Number of outputs and
inputs Method
Mean efficiency degrees
Remarks and other findings
EducationRhodes and South-wick (1988)
64 public and 57 private
universities in US
Panel annual 1971, 1974,
1981
5 outputs5 inputs
Non parametric
About 88% a year
- Private universities have slightly hither efficiency scores, for everyyear considered
RailwaysOum & Yu (1991)
21 railways companies
Annual data 1 output1978-1988
Parametric 1 each year - Limited evidence has been found for a relationship between the share of state in capital and cost efficiency- Positive correlation appears between cost efficiency and the importance of the cantons’s participation in the deficit of firms
Filippini & Maggi (1991)
57 railways under mixed ownership
Annual data1985-1988
1output3 inputs
+2 network characteristics
Non parametric
81% - Tendered services have higher efficiency scores that non-tendered ones.
Productive efficiency comparative studies of public and private firms
21
Is it worth the amount of time?Yes, but with caution
Technical efficiency is just one aspect of efficiency.
Lack of quantitative variables may distort the results.
For education importance of employability.
22
4. Measuring the performance of the public sector as a whole
Ideally:
Data on happiness (average and distribution) with and without social protection or at least on how the welfare state fulfils its objectives: health, education, employment, poverty alleviation, inequality reduction;
Data on inputs.
23
Actually:
Data on indicators of social inclusion (or exclusion);
Data on social spending.
24
Three issues:
Aggregation: DEA or SPI,
Scaling: (0,1) or average or goalposts,
Use of inputs: performance versus
inefficiency.
25
Table 1: Indicators of exclusion. Definition and correlations
Definition
POV At-risk-of-poverty rate
INE Inequality
UNE Long term unemployed
EDU Early school leavers
EXP Life expectancy
Correlation
POV INE UNE EDU EXP
POV 1.000
INE 0.912 1.000
UNE 0.420 0.409 1.000
EDU 0.668 0.782 0.252 1.000
EXP -0.069 -0.098 0.084 -0.203 1.000
Source: The five indicators are taken from the Eurostat database on Laeken indicators (2007).
26
Table 2: HDI normalization and SPI1 - 2004
POV INE UNE EDU EXP SPI1 Rank AT 0.80 0.87 0.93 0.99 0.57 0.83 2 BE 0.60 0.82 0.33 0.89 0.53 0.63 8 DE 0.50 0.72 0.04 0.88 0.58 0.54 10 DK 1.00 0.97 0.96 1.00 0.07 0.80 4 ES 0.10 0.54 0.48 0.25 0.91 0.46 13 FI 1.00 0.95 0.76 0.99 0.00 0.74 6 FR 0.70 0.77 0.37 0.82 0.87 0.70 7 GR 0.10 0.31 0.00 0.79 0.51 0.34 14 IE 0.00 0.56 0.87 0.86 0.35 0.53 11 IT 0.20 0.41 0.35 0.55 1.00 0.50 12 LU 1.00 0.90 0.98 0.86 0.35 0.82 3 NL 0.90 0.82 0.87 0.82 0.54 0.79 5 PT 0.00 0.00 0.57 0.00 0.00 0.11 15 SE 1.00 1.00 0.96 1.00 0.90 0.97 1 UK 0.30 0.49 1.00 0.79 0.47 0.61 9 Mean 0.55 0.68 0.63 0.77 0.51 0.63
27
Difference in shadow prices
SPI1 SPI2
POV -0.02 -0.03
INE -0.05 -0.04
UNE -0.04 -0.05
EDU -0.006 -0.010
EXP 0.06 -0.003
Correlation: 0.9Dependent on irrelevant alternatives.
SPI1 and SPI2
28
DEA with same input:- DEA1: 0.921- DEA2: 0.990
DEA is not invariant to non linear transformation.- DEA3: 0.992
29
Figure 1: DEA1 frontier
q1
q20
C
A
B
D
E
D*
E*
F*
F
30
DEA1 DEA2 DEA3 Scores rank Scores rank Scores rank
AT 0.995 7 0.988 9 0.999 7 BE 0.892 12 0.983 12 0.972 14 DE 0.886 13 0.984 10 0.975 13 DK 1.000 1 1.000 1 1.000 1 ES 0.939 8 0.997 7 0.996 8 FI 1.000 1 1.000 1 1.000 1 FR 0.937 9 0.997 7 0.995 9 GR 0.795 14 0.981 13 0.969 15 IE 0.900 10 0.976 14 0.995 10 IT 1.000 1 1.000 1 1.000 1 LU 1.000 1 1.000 1 1.000 1 NL 0.900 10 0.984 10 0.995 10 PT 0.565 15 0.959 15 0.980 12 SE 1.000 1 1.000 1 1.000 1 UK 1.000 1 1.000 1 1.000 1
Mean 0.921 0.990 0.992
Note: DEA1, DEA2 and DEA3 results correspond to HDI, Afonso et al. and “goalspot” normalization data respectively.
Table 3: DEA efficiency scores. 2004
31
SPI1 SPI2 SPI3 DEA1 DEA2 DEA3 SPI1 1.000 SPI2 0.894 1.000 SPI3 0.959 0.883 1.000 DEA1 0.801 0.643 0.750 1.000 DEA2 0.669 0.517 0.598 0.903 1.000 DEA3 0.583 0.576 0.405 0.679 0.656 1.000
Table 4: Correlations between indexes
32
Measuring performance or efficiency
Problem: weak link between social spending and education, health, unemployment.
Ranking modified
33
DEA1 DEA1 Scores rank Scores rank
AT 0.995 7 AT 0.917 8 BE 0.892 12 BE 0.809 12 DE 0.886 13 DE 0.769 13 DK 1.000 1 DK 0.824 11 ES 0.939 8 ES 1.000 1 FI 1.000 1 FI 0.943 6 FR 0.937 9 FR 0.924 7 GR 0.795 14 GR 0.752 14 IE 0.900 10 IE 1.000 1 IT 1.000 1 IT 0.988 5 LU 1.000 1 LU 1.000 1 NL 0.900 10 NL 0.864 9 PT 0.565 15 PT 0.444 15 SE 1.000 1 SE 1.000 1 UK 1.000 1 UK 0.825 10
Mean 0.921 Mean 0.871
Table 5 DEA efficiency scores without and with social expenditures as input. 2004
34
Race to the bottom?
Test of convergence SPI1 and Malmquist decomposition
35
SEFI
DK
DENL
ATFR
LUBEGR
UK
ITIE
PT
ES
y = -1.2741x + 1.0326
R2 = 0.8024
-1%
0%
1%
2%
3%
4%
5%
6%
7%
8%
9%
0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1,0
SPI1 - 1995
Gro
wth
rat
e o
f S
PI1
(1
995-
2004
)Figure 6: Convergence of SPI1
36
ITES
GR UK
BE
FR DENL
IE
AT
DK
FILUPTSE
y = -0.0862x + 0.0853
R2 = 0.9468
-1%
0%
1%
2%
3%
4%
5%
0,4 0,5 0,6 0,7 0,8 0,9 1,0 1,1
DEA1 1995
Ave
rag
e E
ffci
cie
ncy
ch
ang
e 19
95-
2004
Figure 7: Convergence of DEA1 according to “technical efficiency” change
37
5. Conclusion Yes for efficiency measures when the
production technology is well understood.
Caution when the technology is unclear and environmental variables are missing.
For the welfare states, ranking performance is preferable.
DEA is to be preferred over SPI.
No clear guidelines on the choice of scaling.