Page 22 - PATIENT REGISTRY DATA FOR RESEARCH: A Basic Practical Guide

acceptable (i.e. neither too large nor too small), the sample size of 384 is generally considered

to be sufficient for estimating the prevalence rate for most of the cases.

In contrast to prevalence rate, the sample size required to estimate the population

mean is usually smaller (Bujang et al., 2012; Bujang et al., 2015). However, this may not

hold true if the margin of error selected by the researcher is different, since a smaller margin

of error will always necessitate a bigger sample size. Therefore, it is up to the researcher to

decide how small the margin of error in sample populations they intend to detect, since a

larger sample will be required to detect a smaller margin of error.

For those studies that involve hypothesis testing, the sample size required is

determined by three components, such as type I error (usually fixed at 0.05), power (usually

fixed at 80.0%) and the effect size. The effect size is estimated by using a different formula,

which shall depend on the specific statistical test employed. It is recommended that a sample

size of at least 500 to be required for performing most of the common statistical hypotheses

Bujang et al.,2015). Nevertheless, researchers should always bear in mind that all sample size

calculations will have to take into account all the specific study objectives, in order to ensure

that the power attained by the study will be sufficient to address all the study objectives.

There are numerous guidelines that currently exist in the literature for estimating the

minimum sample sizes required for performing (i) the sensitivity and specificity test, (ii) the

correlation test, (iii) the intra-class correlation coefficient, (iv) kappa agreement, (v) multiple

linear regression and (vi) analysis of covariance, (vii) Cronbach's alpha test, (viii) logistic

regression, (ix) survival analysis (Concato et al., 1995; Bujang et al., 2016; Bujang &

Baharum, 2016; Bujang & Baharum, 2017a; Bujang & Baharum, 2017b; Bujang et al., 2017;

Bujang et al., 2018a; Bujang et al., 2018b).