Start with what you know about the population
If you know the population size — students in a faculty, registered traders in a market — the Yamane formula gives a quick, widely accepted figure at a 5% margin of error. If the population is effectively unknown, use Cochran with an assumed proportion of 0.5, which gives the most conservative number.
Enter both in the Sample Size Calculator and compare. Where the two disagree sharply, your population estimate is doing most of the work and should be justified in the methodology.
Adjust for non-response before you go to the field
A calculated minimum is the number of usable responses, not the number of questionnaires to distribute. With face-to-face administration in Port Harcourt, plan for 10-20% loss; with online distribution, plan for far more.
Divide the calculated sample by your expected response rate to get the distribution target.
Use power analysis when you are comparing groups
Margin-of-error formulas answer 'how precise is my estimate?'. They do not answer 'can I detect a difference between two groups?'. For comparative or experimental designs, use power analysis with a stated effect size, 80% power and a 5% significance level.
Small expected effects need large samples. Discovering that after data collection is not recoverable.
Account for clustering
If you sample whole classes, wards or communities rather than individuals, responses inside each cluster resemble one another. Multiply your sample by the design effect to keep the same precision — with a modest intra-cluster correlation this can raise the requirement by half or more.