Contracting out and quality of road maintenance service in Uganda using

Panel dataPresented by Miriam Katunze,

PhD Student, Lingnan University HKAt the Evidence to enhance inclusive growth conference

04th November 2020

11/04/2020 1

Presentation Outline

• Abstract• Introduction• Scenario setting• Literature Review• Methods• Data• Results• Discussion and Conclusion

11/04/2020 2

Abstract• While it has been established that contracting out brings about cost

savings in public service,• Evidence has been largely associative than dynamic• This study examines the effect of contracting out on the quality of road

maintenance service in Uganda using panel data

• The results reveal a negative effect contrary to many prior studies• A number of reasons are attributed to this finding• Limitations and recommendations for future research are suggested

11/04/2020 3

Introduction• Privatization has popularized contracting out over and above other

public management reform systems. • Contracting out (CO) refers to the use of the private sector in public

service provision

• This due to a myriad of benefits that research has demonstrated • For example in the transport sector CO has led to longer length,

flexibility of delivery (Dempsey, Burton, & Selin, 2016; Meduri & Annamalai, 2013)

• However such compelling evidence lacks dynamism,

11/04/2020 4

• i.e. the associative and or causal benefits/disadvantages of CO over time are largely lacking especially on Africa

• Few assess technical public services like road maintenance (Blom-Hansen, 2003; Pertersen and Houlberg, 2017)

• Prior studies focus on transaction costs as outcome variables with the quality of public service as a control variable.

• Moreover, there are no comparable studies on Africa

• Hence this study examines the effect of CO on the quality of road maintenance service in Uganda using panel data.

11/04/2020 5

Scenario setting• Although not uniform across local governments, CO has been observed

in Uganda• For instance in local markets, bus parks, slaughterhouses, recreation

centers and even roads• Road fund disburses funds for road maintenance through a decentralized

system• CO first seen in 1993- Labor based contracting programme• In that last 20 years funds have oscillated between contracting out and

the use of government own equipment. • Although the latter is currently used, CO is used in some local

governments especially those that get funds from donors.

11/04/2020 6

Literature review• CO is motivated by ownership and competition argument (Alonso et al. 2015;

Bennmarker et al, 2013)

• Cost savings in road maintenance (Blom-Hansen, 2003) but not uniform across municipalities (Petersen and Houlberg, 2016)

• And savings are higher with the first tender and lower with subsequent contracts (Lindholst,et al. 2018)

• Compared to intermunicipal cooperation, savings are fewer with CO if fixed and duration effects are considered (Dilkgraaf & Gradus, 2013)

11/04/2020 7

• Cost savings are high even with sparsely populated municipalities (Soukopva et al, 2017)

• The use of quality of public service as an outcome variable is demonstrated by mixed effects, for instance;

• Positive for first time public transport contracts and negative for subsequent tenders (Mouwen and Reitvel, 2013)

• Positive association when a portion of an education budget is contracted out (Rho, 2013).

11/04/2020 8

Methods• Parameters 𝛼𝛼 and 𝛽𝛽 are estimated by an ordered logit maximum

likelihood random effects model. For each 𝑖𝑖, the log likelihood function is (Wooldrige, 2002):

• ℓ𝑖𝑖 𝛼𝛼,𝛽𝛽 = 1 𝑄𝑄𝑄𝑄𝑖𝑖 = 1 𝑙𝑙𝑙𝑙𝑙𝑙 Φ 𝛼𝛼2𝑥𝑥𝑖𝑖𝛽𝛽 + 1 𝑄𝑄𝑄𝑄𝑖𝑖 = 1 𝑙𝑙𝑙𝑙𝑙𝑙[]

Φ()

𝛼𝛼3 −𝑥𝑥𝑖𝑖𝛽𝛽 − Φ 𝛼𝛼2 − 𝑥𝑥𝑖𝑖𝛽𝛽 + ⋯+ 1 𝑄𝑄𝑄𝑄𝑖𝑖 = 𝐽𝐽 𝑙𝑙𝑙𝑙𝑙𝑙 1 = Φ 𝛼𝛼𝐽𝐽 − 𝑥𝑥𝑖𝑖𝛽𝛽

• 𝑥𝑥= CO (+); number of households/urbanization (+), length of roads in local government-LG (+); kilometers of road maintained (+); commonest transport to reach nearest road/wear and tear (-); frequency of repairs (+); fixed time effects; region and year effects; Cost of maintaining roads (-)

11/04/2020 9

Data • Uganda Household Panel Survey (UNPS)’s community model for

2013/14 and 2015/16. • Service delivery; general community social economic characteristics

• Quality of road service is recoded to fit ordinal regression i.e. 1= good, 2= average, 3= poor

• Main actor responsible for road maintenance not included as an explanatory variable-

• because service quality is likely to be correlated with the main actor or contractor,• concurrent road maintenance by government and contractors

11/04/2020 10

• CO is a binary variable (Y/N)

• Four types of roads are investigated in the dataset but only community access roads are considered for analysis

• Unit of analysis is the community but because LG’s are responsible for the service, I refer to LG’s in the results section instead of communities

• Cost= budget release in Uganda shillings/total road length; transformed to log

• Unbalanced panel of 594 observations.

11/04/2020 11

Results

11/04/2020 12

Table 1: Descriptive statistics 2013-2016. Overall, between and

within variationSource: Author based on UNPS 2013-2016

11/04/2020 13

Variable Mean Std. Dev. Min Max ObservationsQuality Overall 1.87 0.61 1 3 357

Between 0.54 1 3 236Within 0.3 0.87 2.87 1.51

CO Overall 1.67 0.47 1 2 357Between 0.44 1 2 236Within 0.22 1.17 2.17 1.51

Population Overall 351.7 646.64 13 10118 548Between 696.44 17 10118 325Within 240.64 -1798.3 2501.70 1.67

Total road length (km) Overall 90.61 72.55 2 588.19 353Between 68.31 2 413.445 233Within 27.14 -84.13 265.36 1.52

Usual transport to road Overall 1.11 0.31 1 2 358Between 0.3 1 2 237Within 0.13 0.61 1.61 1.51

Frequency of repairs Overall 2.95 1.51 1 5 357Between 1.4 1 5 236Within 0.63 0.95 4.95 1.51

Length of road maintained (km) Overall 27.74 38.55 0 312 352Between 41.3 0 312 232Within 12.94 -51.51 106.99 1.52

Cost (log) Overall 11.86 1.86 -3.71 16.55 289Between 2.07 -3.71 16.55 195Within 0.42 9.88 13.84 1.48

11/04/2020 14

1 23a 4a 5a

CO (Ref= No) -1.20* -1.55** -1.20* -1.55* -1.58**0.02 0.01 0.01 0.02 0.01

Population 0.00 0.00 0.00 0.00 0.000.51 0.66 0.52 0.71 0.51

Total road length (km) 0.01* 0.01* 0.01* 0.01* 0.01*0.03 0.02 0.02 0.01 0.03

Usual Transport to road (ref= Walking) 0.04 0.12 0.04 0.12 0.090.95 0.88 0.96 0.90 0.92

Frequency of repairs (ref=routine manual)Routine-mechanized -0.29 1.53* -0.29 1.53* 1.26+

0.64 0.05 0.61 0.04 0.08Regular manual 0.38 0.65 0.38 0.65 0.64

0.54 0.33 0.57 0.35 0.37Regular-mechanized 0.52 1.53+ 0.52 1.52+ 1.64*

0.43 0.07 0.44 0.06 0.03Other -0.69 0.22 -0.69 0.22 0.50

0.28 0.77 0.33 0.80 0.54Length of road maintained (KMs) -0.02* -0.02+ -0.02** -0.02* -0.01*

0.01 0.05 0.00 0.03 0.02Cost (log) 0.28* 0.3* 0.3* 0.3* 0.25*

0.03 0.04 0.01 0.01 0.03Year fixed effects No yes No Yes yesRegional fixed effects No yes No Yes yescut1 -1.38 -0.16 -1.38 -0.16 -1.31

0.37 0.94 0.29 0.93 0.45cut2 4.66* 6.70** 4.66** 6.70** 5.44*

0.01 0.01 0.00 0.00 0.01sigma2_u 4.51+ 4.23 4.51+ 4.23 4.58+

0.07 0.10 0.06 0.11 0.07Observations 282 282 282 282 282Wald chi2 13.51

9.91b 18.92 39.09 29.72

prob>chi2 0 0 0.04 0.06 0.03

Table 2: Effect of CO on road maintenance quality

Notes: P-values (+ 0.10 * 0.05 ** 0.01 *** 0.001). aModels 3-5 are based on robust

estimates. Model 5 selects a few regions, bDifferent from models 3-5 as LR test vs. ologitmodel: chibar2 is considered. Source: Author

based on UNPS 2013-2016

11/04/2020 15

1 2 3CO (ref=No) -2.54** -2.59** -2.52**

0.00 0.00 0.00Usual Transport to road (ref=Walking)

0.09 0.01 -0.02

0.87 0.99 0.98Population 0.00 0.00 0.00

0.13 0.12 0.13Total road length 0.00 0.00 0.00

0.74 0.80 0.98Length of road maintained 0.00 0.00 0.00

0.87 0.85 0.85Cost (Log) 0.08 0.05 0.08

0.50 0.66 0.47Constant 2.75+ 3.16+ 2.64+

0.09 0.06 0.09Fixed effects Yes Yes Yeslnsig2u -2.60 -2.99 -14.48Constant 0.87 0.94 1.00Observations 594 594 594

Table 3: Robustness check of the effect of CO on road maintenance quality on imputed values (MI estimation)

Notes: P-values (+ 0.10 * 0.05 ** 0.01 *** 0.001). Source: Author based on UNPS 2013-2016. Model 1-Region,

2-Terrain and 3-Year effects. MI estimation could not allow all three

effects in the same specification.

11/04/2020 16

1 2 3 4CO (ref=No) -1.72* -1.55* -1.56* -1.54*

0.01 0.02 0.02 0.02Medium population 0.36

0.45High population 1.05+

0.09Total road length (km) 0.01* 0.01* 0.01* 0.01*

0.01 0.01 0.01 0.01Usual Transport to road (ref= Walking) 0.30 0.12 0.11 0.12

0.74 0.90 0.90 0.89Frequency of repairs (Ref=Routine manual)Routine-mechanized 1.51* 1.53* 1.51* 1.52*

0.04 0.04 0.04 0.04Regular-manual 0.72 0.65 0.68 0.65

0.31 0.35 0.33 0.35Regular-mechanized 1.53+ 1.53+ 1.53+ 1.53+

0.05 0.06 0.05 0.06Other 0.34 0.22 0.24 0.23

0.70 0.80 0.78 0.79Length of road maintained (km) -0.02* -0.02* -0.02* -0.02*

0.03 0.03 0.03 0.03Cost (log) 0.29* 0.3* 0.28* 0.29*

0.02 0.01 0.02 0.01Centered population 0.00

0.71Log population 0.28

0.41Pop 0.00

0.72cut1 -0.85 -1.69 -0.32 -1.55

0.72 0.45 0.91 0.48cut2 6.05* 5.18* 6.51* 5.25*

0.03 0.04 0.04 0.04sigma2_u 4.37 4.23 4.11 4.27

0.11 0.11 0.12 0.11Observations 288 282 282 278Wald chi2 37.29 39.09 37.22 35.41Prob>chi2 0.11 0.06 0.09 0.10

Table 4: Robustness check of the effect of CO on road maintenance quality with transformed population values and capital city values removed from analysis

Notes: P-values (+ 0.10 * 0.05 ** 0.01 *** 0.001). Source: Author

based on UNPS 2013-2016. Model 4 removes values on the

capital (Kampala) from the analysis

Discussion and Conclusion• The main result indicates a negative relationship between CO and quality

of road maintenance service. • followed Blom-Hansen (2003), and Petersen and Houlberg (2017), but

the results are not comparable • economies of scale could not be assessed• prior studies considered costs as a major dependent variable and quality as the

independent variable

• The strength of the paper is the address of selection bias through the addition of fixed time and time invariant effects

• Results collaborate with Rho (2013); Camp and Gaes (2002); O’Toole and Meier (2014); Purse (2009) and Thompson (2011).

11/04/2020 17

• Corroborate prior research on service quality and CO in various fields such as education, prisons, and workers’ compensation claims.

• Several possible explanation for the result:• Contracting may reduce spending which lowers service quality as both

governments and contractors have the tendencies to save costs to profit more from outsourcing

• CO is synonymous with bloated bureaucracy (O'Toole Jr & Meier, 2004; Rho, 2013),

• which affects service quality due to delays in processing or participating parties not honoring contracts due to poor legal enforcement

11/04/2020 18

• CO requires a strong relational element which most governments in developing countries have not mastered (Batley & Mcloughlin, 2010).

• Most contracts are funded through donors, which means an undermining of state capacity, hence giving contractors a leeway to save costs.

• Suggestively, CO may not be suitable for fragile countries (Batley & Mcloughlin, 2010) but could be extended to a regional outlook,

• Uganda is not so much of a fragile country but some of its districts border fragile countries such as the North and the South West (see Regional effects)

11/04/2020 19

• The paper has several limitations:• The service quality variable may not just speak for the survey year as

contracts are usually lengthy,

• Its effect may well go beyond the survey year. • Contractual performance is usually impressive only at the onset (Lindholst et al.

2018a)• Road quality arises from previous years’ maintenance (Petersen & Houlberg,

2016)

• Database lacks control variables for contractors

• Future studies on technical public services should ensure adequate data on transaction costs

11/04/2020 20

Thank you for listening, I welcome comments and suggestions

miriam.katunze@gmail.com/miriamkatunze@ln.hk

11/04/2020 21