Data sources and methods of analysis
2.2 The TDHS data limitations
The major concern of the TDHS survey was to collect a wide range of information concerning the demographic and health status of the survey population. The data were collected by a research team. However, a number of limitations in these data must be considered. Four crucial limitations are particularly relevant to this study.
Firstly, the TDHS is a nationally representative survey providing data on fertility, family planning, and maternal and child health to facilitate national planning, policy making and research on population and health. By contrast, the present study focuses on utilisation of maternal and child health services and child health among children aged 0-59 months in the Northeast region only. The number of cases that meet the criteria for this study was rather small. This limitation might have an effect on the analysis; in particular, the small number of cases will limit the statistical methods to be used. In addition, since all children aged 6-36 months were included in the analysis of nutritional status and those aged 0-59 months in the analysis of diarrhoeal diseases, some characteristics of women and their husbands may be duplicated if they had more than one child. However, the numbers of children who fall in such groups may be quite small, and thus will not have a significant effect on variables under study.
Secondly, the most crucial demographic variable in this, as in many studies, is age. Most studies indicate that Thai women tend to state their age at their next birthday as their current age. The TDHS age data on women and children were
obtained in response to the question 'How old are you?' In addition, respondents were asked whether it was possible to obtain the date of birth from a household register card or birth certificate. Only slightly more than half of the children's ages were reported from a birth certificate. Although birth date, including time of birth, is considered to be important for villagers in the Northeast in order that this information is used to estimate the age at some important events in the life cycle such as age at ordination for men and marriage (Centre for Isan Monks, 1990: 3), the accuracy of the reported birthdate and age relies on the respondents' memories.
Thirdly, the measurements of body weight and height of the children are usually prone to measurement error, particularly in young infants (Bailey, 1993: 10), as a result of errors made by field workers on errors in the instruments. Field workers for the TDHS were given extensive training, and each fieldworker's measurements were standardised with those of the supervisor (Chayovan et al., 1988: 111). Height measuring boards with a fixed headboard and a movable footboard were used to measure the recumbent length of children, who were also weighed wearing minimum clothing on a portable and durable lightweight hanging scale. Children under one year old were put in an infant's basket and those one year or older were held by weighing pants when weighing (Suntikitrungruang and
Nokyoongthong, 1989: 60). Although these anthropometric data allow an
examination of a wide range of nutritional status indicators, since the data are cross- sectional, this does not take into account seasonal variations in malnutrition which are pronounced in the Northeast (Kotchabhakdi et al., 1987; Pongpaew et al., 1990).
Fourthly, the morbidity data on the occurrence of diarrhoea were obtained retrospectively. The data were reported by the mothers of the children, and recorded by a non-medical person. Even with a short recall period of 24 hours and two weeks, and extensive training of interviewers, the chances of under-reporting or mis-reporting cannot be discounted. Over-reporting, on the other hand, is unlikely, so that measures have an inherent negative bias. Apart from memory lapse,
perceptions about and attitudes toward disease influence reporting of the occurrence of disease. For example, if people perceive a certain disease to be a common or minor ailment, when the disease occurs, it may not be considered as an incidence of disease (Wadsworth et al., 1971: 93).
The study is an example of perception of infant diarrhoea in rural Northeast Thailand. Thongkrajai et al. (1990: 774) indicated distinct differences in the definitions and terms used for diarrhoea according to a number of criteria. The conventional term thongsia was used only if frequent loose stools occurred in adults or children. Among infants or children under one year of age, the same symptoms were called sou. M ost mothers believed that sou was normal for infants, who are believed to pass frequent stools with changing stages in body growth and development. The TDHS, however, used the conventional term thongsia, although sometimes, but not always, the local term for diarrhoea (which differs among regions), was used. When the conventional term thongsia was used, information on sou would not have been obtained, and consequently the morbidity rate may have been grossly under-reported.
The extent of under-reporting due to memory error or intentional and unintentional failure to mention disease may not have been uniform among the mothers of different socio-economic groups, and as such can pose a serious limitation in studying morbidity differentials. In obtaining morbidity data an attempt was made by the TDHS to reduce field staff biases through extensive training. A field check of questionnaires was also made by supervisors (Chayovan et al., 1988:
111).
The overall data quality is discussed in the TDHS 1987 report (Chayovan et al., 1988: 157-165). The errors which might occur during the field survey (non sampling errors) are difficult to avoid but these problems were handled carefully during the design and implementation of the survey, thus minimising these problems as far as possible. The sampling errors were evaluated statistically and measured by means of the standard error of some selected variables. The conclusion of the
examination of the measures is that the data for the country as a whole were reliable, as the sampling errors for selected variables were small. Similar results are also found for the Northeast region (Chayovan et al., 1988: 159).