Round table discussion on research design, statistical aspects and data collection.

Abstract

Sample size determination, data collection, and suckling response are protocol discussed by Harrison, Hiorns, and Mascie-Taylor. The advice is to use a representative random sample when little is known, and a matched sample when some information is available and the interest is in particular sections or aspects of groups in the population. The optimum design for a known linear phenomena is to select fertility examples from the extremes (1 child vs. 10 children), however, if the pattern is U-shaped or nonlinear, it would be inappropriate. Bias in cluster sampling will occur if it is restrictive and will be identifiable if it is producing an effect. A random sample is not always necessary. Sample size determination is related to the statistics and tests to be used in the analysis and the primary hypothesis and practical cost. Validity is a convincing argument for the research choices made. The number of comparisons required is greater in determining the probability of rejecting the null hypotheses. Types of recorded variables (continuous, binary, categorical) must be stated a priori. Null hypotheses and alternative hypotheses must be noted preliminary to the study. Secondary analysis or "go-fishing" analyses are performed when new hypotheses are considered in the course of the study. Negative results should be reported but frequently are not, particularly if funding is at risk. In descriptive studies, the larger the sample, the greater the accuracy in describing the population. In comparative studies, determining that the differences are due to the exposition variable means eliminating confounding variables. This is compounded with a control group. Time is another important variable. It is better to use a few selected variables than too many. The minimum is the nutritional status of the mother and child, body composition of the mother, maternal food intake and weaning foods, energy expenditure and breast-feeding patterns, fertility questions, and urine or saliva samples. In a longitudinal study, observations should be made at regular intervals. WHO has set up a protocol for assessing breast-feeding patterns. Menstrual status is the easiest to obtain. Hormonal profiles should be taken with easy and noninvasive methods. To obtain progesterone data, collect samples once a week. Suckling response can be measured. Protocol is discussed for a study relating maternal anthropometry to pregnancy outcome.