Mama Mdogo
JF-Expert Member
- Nov 21, 2007
- 2,961
- 2,138
Mzee, with all due respect, I beg to differ. I am not a statistician, however, the little I know tells me that you need to define your sampling UNIT (the smallest population element where the measurements are going to be made) before you conduct your research. In this case, it is an individual person eligible for voting (more precisely also registered for voting and a higher probability of casting a vote!). The list of these units forms what is called a sampling frame. If the population is homogeniousy distributed (e.g. same age, education level, marital status, gender, etc, etc) a random sample of these units is adequate to be a representative sample. The question of how many units, is determined by the level of confidence (conventionally accepted at 95%, 99%, or 99.9%), variation (variance or Standard deviation) and the level of precision desired (OR acceptable error margin) conventionally kept at 0.05. THIS IS WHEN YOU USE A SINGLE STAGE SAMPLING.
In the citizen newspaper, the REDET sampling strategy involved a multistage (3-stage) sampling with 26 regions, 52 districts, 102 villages, 2,600 people interviewed! This is not a homogeneous population anymore. For this design, they needed to include in the formula, the design effect (3 stage stratification) and a factor called the intraclass correlation (ICC) which quantifies the extent of between-group variation (or within group homogeneity). The rough estimate I made suggests the they needed at least 91,800 units for this polls. Mind you, I have not included the possible confounding affects of such factors as age, education level, marital status, gender, etc.
Conclusion: What the REDET findings tells us NOTHING regarding the coming election and that the study is null and void. At least they should have indicated confidence intervals to try and convince us otherwise.
Cheers
Even if REDET would have considered the needed at least 91,800 units for this polls and the indicated confidence and precision, at the end of the day, their data needed to be weighted to accommodate differences across regions or districts. The study is thus null and void.