CHAPTER 2 Conceptual framework, data and methodology
2.3. Quality of data
2.3.1. Philippine census data
The National Statistics Office, formerly the National Census and Statistics Office, has conducted post-enumeration surveys (PES) to evaluate the 1970, 1975 and 1980 censuses. The 1970 and the 1975 PES were undertaken on a national scale while the 1980 PES was confined to Metro Manila only. However, the results of these surveys were not released due to problems arising from lack of independence between these censuses and the corresponding post-enumeration surveys.
To ensure that the same problems were not repeated in the 1990 post enumeration survey, the National Statistics Office commissioned the Philippine Social Science Council (PSSC) to carry out the survey. The 1990 post-enumeration survey, which was referred to as the 1990 Census Evaluation Survey (CES), covered a nationally representative sample of 16,219 households. The PSSC took the responsibility of implementing the 1990 CES. However, in some aspects of the CES, employees of the National Statistics Office who were not directly involved in the 1990 census provided the necessary assistance to the PSSC.
Based on the initial results of the 1990 CES, estimates of the net coverage error rate, which is the difference between the omission rate and the erroneous enumeration rate, in the 1990 census, were derived. These estimates range from an over-count of 0.36 per cent to an under-count of 6.57 per cent. An under-count of close to 2 per cent was considered the most plausible (National Statistics Office (NSO), 1992c: 3). The under-count is believed to be neither selective of, nor biased against, any population group. Accordingly, this does not affect the present study, which deals primarily with trends and patterns, that is, percentage distributions of family households and family household population, rather than with absolute numbers.
Because the results of the post-enumeration surveys for the censuses of 1970 and 1980 are not available, the coverage error in these censuses cannot be measured directly. Thus, the extent of coverage error in the 1970 and 1980 censuses is evaluated
by comparing the average annual growth rates during the inter-censal periods 1970- 1975 and 1975-1980 with the rates of growth in preceding inter-censal periods.
Except for such unusual events as war, epidemics, disasters and massive migration resulting in either a substantial gain or loss in the number of people, the rate of growth of the population for the entire country as well as for each region and province in the country is expected to change gradually from one inter-censal period to another. The average yearly rate of growth of the population of the Philippines between censuses from 1903 to 1990 has been shown in Table 1.1.
In their assessment of the coverage of censuses of population from 1903 to 1975, Mijares and Nazaret (1978: 338-344) suggested that abrupt changes in the rate of growth that took place at some points within the 1903-1975 period can be explained by the occurrence of an unusual event. For instance, during the period 1918-1939, the rate of growth accelerated due to the control of epidemic diseases, and hence the drop in death rates, brought about mainly by the introduction of modern medical and sanitation measures. Another factor, according to these authors, that could have accounted for the acceleration of the growth rate is the good coverage of the 1939 census. Regarded as the most complete and accurate census ever taken in the Philippines, the 1939 census was carried out when all areas of the country were in almost perfect peace and order (Mijares and Nazaret: 1978: 339). In contrast, the decline in the rate, from 2.2 per cent per annum during the period 1918-1939, to 1.9 per cent annually between 1939 and 1948, was attributed to a number of factors obtaining in the 1948 census: internal disorder, destruction by war, the taking of the census during the typhoon season when travel and communication are most difficult, errors committed by the census enumerators who were trained for three days only and therefore were not able to master the intricacies of gathering data, and inaccurate data furnished by the respondents for reasons such as impatience, hostility and fear of being enumerated (Mijares and Nazaret, 1978: 339).
In the censuses from 1960 to 1980, the coverage for the country as a whole can be considered reliable. The rate of growth of the national population declined gradually from 3.0 per cent annually during the 1960-1970 period to 2.7 per cent yearly during the 1970-1980 period. At the regional level, however, there was an indication of under enumeration in two regions, namely Western Mindanao and Central Mindanao, during the 1975 census (Cabigon, 1990: 106). These two regions had unusually slow growth rates during the period 1970-1975 (1.8 per cent and 1.3 per cent annually, respectively). This problem concerning the 1975 census coverage will not affect the present study as it does not utilise data from the 1975 census.
b. Quality of census age-sex distribution data
Age and sex are the two demographic characteristics considered most basic. Information on these characteristics are collected in all censuses and sample surveys. They are easy to report and thus theoretically the data on age and sex should have a higher level of accuracy than for any other items in the survey schedule. The quality of age and sex data can therefore be considered as an overall index of the quality of the content or the quality of the other data collected in a census.
The most common source of error in age data from censuses is the tendency of respondents to report or of enumerators to record ages ending in certain digits at the expense of other ages. This is called age heaping, age preference or digit preference (Shryock et al., 1976: 115). One popular measure of the extent of age heaping, which at the same time is a good indicator of the country's literacy and knowledge of age, is Myers' index. The theoretical range of this summary index is 0 to 90. A value of 0 represents no heaping, while a value of 90 means that all ages were reported at a single digit (Shryock et al., 1976: 118).
The Myers' index of age preference calculated for the Philippine male and female population based on the age data recorded in the censuses of 1970, 1980 and 1990 is shown in Table 2.1. A positive index corresponding a particular terminal digit suggests that the digit is overly selected or preferred while a negative index indicates
avoidance of that digit. The data in Table 2.1 suggest that in the 1970, 1980 and 1990
censuses there is a tendency to report ages ending in 0 and 5. A large decline in the
Myers' index, from 7.92 for both sexes in the 1970 census to 2.61 in the 1980 census,
indicates more reliable age data in the 1980 census. The corresponding figure for the
1990 census was 2.32, suggesting further improvement in the quality of age data.
The comparison made by Cabigon (1990: 113) of the Myers' index calculated
using the Philippine data for the census years 1960, 1970, 1975 and 1980 with that for
Indonesia and for broad regions of selected developing countries, has shown that age
heaping in the Philippine census data is not as pronounced as that found in the 1971 and
1980 census data of Indonesia and in the data of some African countries.
Table 2.1. Myers' index of age heaping by sex, census years, Philippines: 1970, 1980 and 1990
1970 1980 1990
Ending
digit of Total Males Females Total Males Females Total Males Females
age a 0 4.46 4.29 4.62 1.78 1.71 1.84 1.51 1.54 1.48 1 -2.31 -2.17 -2.44 -0.75 -0.74 -0.75 -0.53 -0.50 -0.57 2 0.24 0.45 0.03 -0.19 -0.07 -0.32 -0.23 -0.17 -0.30 3 -1.19 -1.17 -1.21 -0.56 -0.54 -0.58 -0.57 -0.59 -0.55 4 -1.26 -1.28 -1.25 -0.28 -0.33 -0.23 -0.16 -0.21 -0.12 5 2.03 1.97 2.10 0.83 0.85 0.81 0.65 0.66 0.65 6 -0.78 -0.89 -0.68 -0.56 -0.59 -0.54 -0.54 -0.57 -0.52 7 -0.91 -0.75 -1.08 -0.03 0.03 -0.09 0.05 0.10 0.00 8 1.20 1.06 1.33 -0.23 -0.27 -0.18 -0.28 -0.33 -0.22 9 -1.47 -1.52 -1.42 0.00 -0.05 0.04 0.11 0.07 0.15 Myers' index 7.92 7.77 8.07 2.61 2.59 2.69 2.32 2.36 2.28
Source: Computed using data from 1970, 1980 and 1990 censuses of population of the Philippines.
c. Content errors in the 1990 census
In the 1990 Census Evaluation Survey (CES), selected questions pertaining to the characteristics of the household population and of housing units were asked. The responses to these questions were compared with the responses to the same population and housing items in the 1990 census, and the results showed a high level of agreement between the two sets of responses with regard to the following demographic characteristics: sex (98.4 per cent), marital status (96.8 per cent), and relationship to household head (94 per cent) (NSO, 1992c: 4). A lower level of agreement was observed for age (86.3 per cent). The information about the relationship to household head, age, sex, and marital status of each household member was used to generate the variables on household type or structure and on family life cycle state. Since the level of agreement between the responses in the census and in the CES was high for each of the aforementioned variables, it can therefore be maintained that the data on household type and family life cycle state are highly reliable. The data on household size are reliable as well. As mentioned earlier, the coverage of the 1970, 1980 and 1990 censuses can be considered reliable.