NSCB - Annual Per Capita Poverty Thresholds by Province, 2003

Some Notes on the Computation
of the Final Provincial Poverty Estimates for 2003

The final poverty estimates for 2003 were computed based on the final results of the 2003 FIES released by the National Statistics Office (NSO) in January 2006. Following are some clarifications on two issues that have so far been raised concerning these estimates.

Level of disaggregation: Regional versus provincial estimates

Preliminary 2003 poverty estimates released by the NSCB in January 2005 were only down to the regional level due to questions on the reliability of the initial provincial estimates. Users, however, demand for provincial level estimates since regional estimates are not very useful .

It must be recalled that before 2003, official poverty statistics used to be available only at the regional level. With the decentralization of government administration, the need for a more disaggregated level of poverty estimates grew. In particular, implementers of the Kapit Bisig Laban sa Kahirapan (KALAHI), the national government's framework program for a focused, accelerated, convergent, and expanded strategy to reduce poverty, was in urgent need to identify the target provinces for prioritizing program beneficiaries. These led to the issuance of NSCB Resolution No. 1, Series of 2003, on 15 January 2003, approving the methodology for the compilation of official provincial poverty statistics. However, the development of the provincial poverty estimation methodology was a long and deliberate process. An evaluation of the soundness of the methodology was done through various computational exercises undertaken by the NSCB Technical Staff, results of which were presented in several consultative meetings. The initial provincial poverty estimates released by the NSCB in compliance with the Resolution became the basis for identifying the 44 poorest provinces initially covered by KALAHI.

The fact that the design of the FIES was not intended for generating provincial level estimates influenced the members of the Technical Committee on Poverty Statistics (TC-PovStat)¹ to take on different views as to whether final poverty estimates for 2003 should be released down to the provincial level. Thus, a referendum was conducted from 16 to 20 May 2006 among the members to decide on the issue. Results of the referendum showed eight members, including the TC Chair, favoring the release of the final poverty estimates for 2003 down to the provincial level and seven favoring the release of the final estimates only at the regional and national levels, with five members failing to respond to the referendum by the set deadline.

Members who favored the release of only regional estimates maintained that the FIES was not designed to generate reliable provincial level estimates and called attention to the high coefficients of variation² (CVs) of the provincial poverty estimates. It was further commented that “ releasing provincial estimates when many of these statistics are not "reliable" will potentially lead to misdirected policy actions (e.g. identification of bottom 40 provinces) that will lead to a lot of leakages in program interventions .”

On the other hand, those who favored the release of provincial estimates did so primarily in response to the users' demand for these data and to maintain the relevance of the official poverty statistics. It was remarked that even developed countries like Australia and the U.S. have released to the public statistics with CVs higher than 10%. Higher CVs do not necessarily mean that the data should be suppressed or not reported. It was, however, emphasized that when estimates with high CVs are released to the public, it is important to ensure that the users know the limitations of the estimates.

The issue, together with the results of the referendum, were presented to the NSCB Executive Board for decision in its meeting on 22 May 2006. Three of the five members who failed to respond to the referendum by the set deadline indicated their positions during the Board meeting itself - two members coming from the National Statistics Office (NSO) favored the release of regional estimates, while the member from the Trade Union Congress of the Philippines (TUCP) favored the release of provincial estimates – bringing the count to nine in favor of releasing only regional estimates and an equal number in favor of releasing provincial estimates .

During the same meeting, the Chairman of the Board expressed his view that regional estimates would not be very useful since regions in the country are not political units (except for ARMM) but merely administrative structures to facilitate the administration of government services. After due deliberation and discussion, the other members of the Board did not express objections to releasing the 2003 final poverty estimates down to the provincial level.

In consideration of the above and to serve the statistical requirements of policy makers and program implementers at the provincial level and thus, maintain relevance of statistics released by the Philippine Statistical System (PSS), the Board approved the release of provincial poverty estimates with the condition that the estimates would be released with accompanying measures of reliability (CVs) to properly caution/inform the public on the degree of precision of the estimates. The Board likewise approved the TC-PovStat recommendations on the refinements in the methodology, namely, (a) the use of exact band in determining the reference population for the food expenditure to total basic expenditure (FE/TBE) ratio; and (b) for the five provinces with no sample urban barangays, u se of provincial rural food threshold as the estimate of the overall provincial food threshold instead of the weighted average of the urban and rural provincial food thresholds as previously done for the preliminary estimates.

CVs and confidence intervals of provincial poverty estimates

Due to the limited sample size of the FIES (from which income and expenditure data were sourced) arising from the government’s budgetary constraints, users are advised to carefully take note of the degree of precision (CV and confidence interval) of each provincial estimate when using the statistics for analysis.

Table 1 shows the distribution of the CVs of provincial poverty incidence estimates for families for 2000 and 2003, while Table 2 presents the provinces with the highest CVs for 2003.

Table 1: CVs of provincial poverty incidence estimates for families

Coefficient of Variation (CV), %	No. of Provinces/Cities
Coefficient of Variation (CV), %	2000	2003
<10	28	28
10-15	32	33
15-20	17	13
20-30	4	6
30-40	1	2
40-50	0	1
>50	0	1

Table 2: Provinces with the highest CVs of 2003 provincial poverty incidence
estimates for families

Province	Coefficient of Variation
Province	2000	2003
Aurora	9.8	59.0
Isabela City	a	45.1
Quirino	16.7	31.2
Apayao	15.9	30.3
Nueva Vizcaya	28.8	27.4
1st District - NCR	21.6	25.9
Bataan	16.8	22.2
Ifugao	8.0	21.9
Camiguin	10.5	21.7
Tawi-tawi	10.2	20.0

a – Isabela City was part of Basilan in 2000.

Although some theoreticians consider 10% as an acceptable level for the CV, official statisticians have not reached an agreement on the issue and enlightened users have used statistics with CVs higher than 10%. In this regard, a noted survey statistician, Prof. Graham Kalton (The Survey Statistician, 2002) says, “…estimates of any quality add to the information available to users. Rather than protecting users from themselves by suppressing estimates that producers judge to be of inadequate quality, I believe that producers should provide users with the estimates together with the information that is needed for users to assess the accuracy for themselves…One widely used form of suppression is to suppress estimates with coefficients of variation of greater than, say, 30 percent…”

The issue on acceptable levels of CV was also raised during a forum held at the Asian Development Bank (ADB) in 2003 where the Prof. Kalton himself told the participants that results of his study on small area estimates of school-age children in poverty in the US were nevertheless used by the government notwithstanding the high CVs for some school districts, because they were still the best available and the government had to have some basis for the allocation of funds.

In addition to the CVs, the NSCB was also required to release confidence intervals of poverty estimates. Confidence intervals, which provide another means of judging the reliability of the estimates, give an estimated range of values that is likely to include an unknown population parameter, which, in this case, is the true poverty incidence, subsistence incidence, or magnitude of poor/food poor. By adjusting the size of the confidence interval, it is possible to specify what proportion of the confidence intervals will include the true value of the population parameter. Confidence intervals are usually calculated at the following percentage/confidence levels: 90%, 95% and 99%. Confidence limits (upper limit and lower limit) are the values that define the range of a confidence interval. In the tables constructed on poverty estimates, 90% confidence intervals were derived.

To illustrate how to interpret confidence intervals, data on poverty incidence with corresponding confidence intervals for the ten poorest provinces in 2003 are presented in Table 3.

Table 3: Ten poorest provinces in 2003 in terms of poverty incidence among families

Province	Poverty Incidence among Families
	Estimates (%)			Coefficient of Variation		90% Confidence Interval
	Estimates (%)			Coefficient of Variation		2000 Revised		2003
	2000 Revised	2003	Inc/Dec	2000	2003	Lower Limit	Upper Limit	Lower Limit	Upper Limit
Zamboanga del Norte	47.0	64.6	17.5	12.0	6.5	37.7	56.3	57.6	71.5
Maguindanao	59.3	60.4	1.1	7.0	7.0	52.5	66.1	53.4	67.4
Masbate	61.3	55.9	(5.4)	5.8	5.7	55.5	67.2	50.7	61.2
Surigao del Norte	42.6	54.5	11.9	11.0	8.1	34.9	50.2	47.2	61.7
Agusan del Sur	52.3	52.8	0.5	10.4	8.8	43.3	61.2	45.2	60.4
Surigao del Sur	38.4	48.6	10.2	8.7	7.2	32.9	43.9	42.8	54.4
Misamis Occidental	46.8	48.1	1.2	7.8	11.7	40.9	52.8	38.8	57.3
Mt. Province	48.4	46.7	(1.7)	13.3	13.0	37.8	59.0	36.7	56.6
Biliran	33.3	46.5	13.3	18.3	13.4	23.3	43.3	36.3	56.8
Lanao del Norte	49.3	46.5	(2.9)	6.7	7.9	43.9	54.8	40.5	52.5

Poverty incidence estimate for Zamboanga del Norte, which was the highest at 64.6 percent, had a CV of 6.5 percent. Thus, we can be 90 percent confident that the true poverty incidence in the province in 2003 lies between 57.6 and 71.5 percent. It must be noted that while the point estimate of poverty incidence in Zamboanga del Norte, which is 64.6 percent, is higher than that of Maguindanao, at 60.4 percent, we can not be absolutely sure that the true poverty incidence in Zamboanga del Norte is higher than that of Maguindanao. This is supported by the overlapping confidence intervals for the two provinces, that is from 57.6 to 71.5 and from 53.4 to 67.4, respectively.

__________________

¹The TC-PovStat has a multi-sectoral representation consisting of noted experts in the area of poverty statistics. One of the Committee’s main tasks is the provision of expert advice in the development of an efficient system of poverty estimation and monitoring

²The coefficient of variation (CV) is the most widely used measure of precision/accuracy of estimates.


E-mail the webmaster	Terms of Use	Home • Top of Page
	1997-2008, National Statistical Coordination Board Makati City, Philippines