Public Expenditure and Service Delivery Survey 2002
A survey of 220 schools
Papua New Guinea
This survey is part of a multi-country pilot study which combines surveys of primary schools with household and other micro surveys to assess service delivery systems in education, measure performance, and establish a baseline for examining the impact of policy and institutional reforms over time.
Work on the PESD project was launched in late 2001 as part of the World Bank’s analytical work on poverty in PNG. The project was launched in close consultation with the Government of PNG and AusAID.8 Work on the PESD survey started in early 2002.
The survey operation itself was implemented by the Education Department of the National Research Institute (NRI) in Port Moresby.
Kind of data
Sample survey data [ssd]
Final datasets, edited.
Relationships not established
The PESD survey covered 214 schools in 19 districts across 8 provinces --Counting NCD as a province-- out of a total of 20 in the country, with two provinces selected in each of the four main regions.
The following provinces were covered:
- Southern (Papua) region: Gulf; National Capital District (NCD)
- Highlands region: Enga; Eastern Highlands
- Momase region: West Sepik (Sandaun); Morobe
- Islands region: West New Britain; East New Britain
These provinces cover a wide spectrum both in terms of poverty levels and educational development. They range from the relatively rich (NCD and Gulf with headcounts of 19 and 28%) to the poor Sandaun (headcount of over 60%), from the well-educated (NCD and East New Britain with adult literacy rates of 84 and 74%) to poorly-educated (Enga and Eastern Highlands with adult literacy rates of 26 and 38%), from those with high primary enrolment (NCD and ENB) to those with low enrolment (Enga, Gulf and Sandaun), from those with high grade 1-8 retention rates (NCD with 79%) to those with low retention rates (Eastern Highlands and Sandaun with just above 20%).
Producers and sponsors
National Research Institute, Port Moresby and Deon Filmer (World Bank)
Three districts were randomly selected within provinces with probability proportional to the number of schools in the district. In two of the provinces, viz. Gulf and West New Britain, that only had two districts, both were selected. Ten schools were then selected randomly within each district. In NCD, which does not have districts but is organized by wards/census enumeration areas, 30 schools were randomly selected.
The original sample included 220 schools. Many of the schools in the original sample could not be covered for a variety of reasons. In these cases, replacement schools (randomly selected from the same district) were used. A special effort was made to ensure coverage of remote schools. In particular, some sites were revisited later to cover schools that could not be surveyed during the first attempt due to logistical difficulties. The schools are widely dispersed throughout the country.
The PESD schools are further classified by the level of poverty and remoteness. The level of poverty is measured by the estimated poverty rate for the LLG where the school is located, and the remoteness index is based on a composite measure of distance and travel time from the school to a range of facilities. The PESD sample of schools is well distributed across the remoteness and poverty spectrum. (For further details on the measures of poverty and remoteness, see Annexes 2 and 3 of the survey report.) Also, while poverty rate and the remoteness indices are significantly correlated across the PESD sample, these attributes are not collinear. The weighted correlation coefficient is 0.15, while the unweighted correlation is 0.27, both statistically significant at the 5% level or better.
The sampling weights reflect the probability of a school being selected from all the schools in a given province. The results of the calculations described here are presented in Table A1.1 in the survey report.
In order for a given school to be selected into the sample, two random events must transpire. Its district must first be selected, and then the school itself must be chosen from all of the schools in the district. So the overall probability of selection is simply the product of the probabilities of each event occurring. Defining a school Si, in district Di and province Pi, we can write:
Districts in Gulf, West New Britain and NCD were automatically selected, and so have a selection probability of one. Three districts were selected from each of the remaining provinces using PPS sampling. This procedure defines the probability of a district being selected in any draw as the number of schools in the district divided by the number of schools in the province, so the overall probability of selection is three times this ratio:
P(Di selected+ = 3 * (number of schools id Di / (number of schools id Pi)
The calculated probabilities of selection for each district are listed in column (c). In East New Britain, two districts (Gazelle and Pomio) were large enough to be selected twice, so the calculated probabilities for these districts were greater than one. We set these probabilities equal to one, and redistribute the excess probability equally between the other two districts.
A Monte Carlo simulation produced empirical estimates of the probabilities which are extremely close to the theoretical results. These estimates are reported in Appendix 1 of the survey report.
Probability of a school being selected
Each school in a selected district has a probability of selection equal to the number of schools selected from the district, divided by the total number of schools in the district:
P(Si selected | Di selected) = number from selected schools of Di / number of schools in Di)
The probabilities of each school being selected are reported in Appendix 1 of the survey report.
Overall probability of selection
The overall probability of selection, reported in column (f), is the product of columns (c) and (e). Column (g) reports expansion factors for each school, which are simply the inverse of the overall probabilities. These give the number of schools in the province represented by each selected school. (The sum of expansion factors for all selected schools in a province should, by definition, equal the total number of schools in that province. Because of the adjustment to the weights for ENB schools described earlier, the expansion factors for ENB schools sum to slightly more than the total 146 schools in the province. We therefore scale the expansion factors for ENB down slightly so they sum to 146.)
The estimated weights are on average greater than one, so the sum of the weights across schools exceeds the number of schools in the survey. To correct for this, the expansion factors were scaled down by a common factor. This also forces the average normalized weight across all schools to be one. The normalized weights and expansion factors are given in Appendix 1 of the survey report.
Dates of collection
Mode of data collection
The survey used a series of instruments for collecting data at different levels. These included:
Instruments at the school level:
- School survey – the main instrument (S1)
- Grade 5 teacher survey (S2)
- Board of Management survey (S3)
- Parent survey (S4)
Instruments at the district/provincial level:
- District Education Administrator (DEA) survey (D2)
- Provincial Education Adviser (PEA) survey (P1)
An instrument for health centers:
- Health facility survey (H1)
These instruments were used to collect data on a range of topics including: characteristics of the head teacher, teachers, characteristics of schools, inspectors, BOM, parents, school finances, classroom environment, teacher activity, resources for teaching, community-school interaction, organization and structure of DEA/PEA offices, District and Provincial Education Boards, budget process, school fee subsidy and other sources of funding, and roles and responsibilities in education.
The health facility survey was not intended to be a full service delivery survey in order to keep the field operations and costs within manageable limits. It was added as a rider to the school survey. Health facilities that could be reached within 20 minutes from the sample schools were covered. Thus, as against a sample of 214 schools, the survey covered 117 health facilities. A short instrument collected information on how often the facilities were open, the presence of staff, and the availability of key medicines. Table 2.2 in the survey report gives details of PESD sample coverage by instrument, province and district.
Use of the dataset must be acknowledged using a citation which would include:
- the Identification of the Primary Investigator
- the title of the survey (including acronym and year of implementation)
- the survey reference number
- the source and date of download
National Research Institute, Port Moresby and Deon Filmer(World Bank). Public Expenditure and Service Delivery Survey (PESD) 2002. Ref. PNG_2002_PESD_v01_M. Dataset downloaded from www.microdata.worldbank.org on [date].
Disclaimer and copyrights
The user of the data acknowledges that the original collector of the data, the authorized distributor of the data, and the relevant funding agency bear no responsibility for use of the data or for interpretations or inferences based upon such uses.