The accuracy of poverty statistics

The poverty problem is being solved, according to a government adviser, who observes that the shopping malls are full, and many new cars are on the roads. Does this portend a new poverty indicator that combines a shopping density index with the sales of new cars? If its convenience-sampling works, that would be a great technical advantage.

Monitoring poverty on the basis of mall shopping. But it won’t be easy for an index derived from mall shopping density and new-car sales—let’s call it Mall Shopping Poverty (MSP)—to replace the present official indicator of the Philippine Statistics Authority (PSA), which is defined by an income line, applied to a periodic family income survey. In the first place, MSP needs a quantitative link to the PSA proportion of poor families, to show the administration’s pace toward its single-digit poverty target. Creating the link will need historical data with the base MSP and PSA numbers matched.

PSA poverty measurement is scheduled for 2023, 2025, and 2027. The 2023 report is half done as of now; the years 2024 and 2026 will be blank. But an MSP, being so simple, could surely be measured much more often—even quarterly, to match the Self-Rated Poverty (SRP) reports of Social Weather Stations (SWS). As a free-market economist, and cofounder of the Foundation for Economic Freedom, I believe that market competition, including competition in research, promotes quality in the economy.

Secondly, MSP should demonstrate its geographical correlation with PSA poverty, which is presently being done for each province and highly urbanized city. It follows that MSP should be monitored with similar detail on the ground. Observing activities in malls and car dealerships nationwide would be pleasant work for the monitoring staff.

Statistics is the science of inferring the characteristics of a population on the basis of a sample, or a small part, of the population. To observe the entire population—meaning each and every person, or else family, in a country—is called a census. However, a census is simply too costly for any institution, even the government, to do on a regular basis; usually it is done only once every 10 years. Aside from censuses, a relatively small sample of the population, drawn scientifically, will have to be surveyed instead, to represent the population.

However, there is one very important exception to the above, in a democratic country; it is the activity of voting in elections and referenda, when every single person’s vote must be counted, and thus resources allocated for the purpose. The results of elections have long served to validate the results of sample surveys of popular intentions (when done prior to voting) and popular will (when done after exiting the polling booths but before the votes are counted)—provided, of course, that the counting is honest. The success of scientific sample surveys in predicting democratic elections is a backstop of statistics.

For the multitude of human behaviors, beliefs, values, attitudes, and other circumstances—including poverty—that do not have a mechanism for fully counting everyone, the way of validating the results of one sample survey is by comparison with other sample surveys which are independent and scientific. The findings of a set of scientific surveys will cluster together; the larger the set, the tighter the cluster. More competition is welcome, because it benefits the credibility of the research community.

SRP has changed significantly, even over a single quarter, over time. Since 1983, SWS has surveyed SRP 142 times; semi-annually in 1986-1991, and quarterly in 1992-2023. Of the resulting 141 time changes in the poverty rate, 37 percent were within the error margin of plus or minus 3 percent, i.e., statistically constant, 33 percent were significant falls, and 30 percent were significant rises. With falls outnumbering rises, the long-term trend in SRP was slightly downward—much more plentiful, but compatible, with the shorter PSA trend in income-poverty in the time when the data overlap.

Of the 33 percent falls, 18 percent were small (3 to 5.9 points), 12 percent were medium (6 to 8.9 points), and 3 percent were large (at least 9 points).

Of the 30 percent rises, 14 were small (3 to 5.9 points), 9 were medium (6 to 8.9 points), and 6 were large (at least 9 points).

The discovery that the changes in SRP have been frequent, as well as large, shows that the frequency of monitoring it has been quite justified.

What historical pattern might have been exhibited by MSP, if it had been formalized and surveyed for some time? Would it agree with the patterns of the PSA and SWS series of data?

—————-

Contact: [email protected].