· Sample size
– Prevalence of current tobacco users (po) = 50% = 0.5
– Level of significance (α) = 5%
– Z1-α/2 = 1.96
– Margin of error (d) = 5%
– Design effect (DE) = 2
– Using formula, Sample size (n) = [Z2 { po (1- po)}/ d2 ]*DE
= [(1.96)2 {0.5 (1-0.5)}/(0.05)2]*2
= 768
– Multiply by number of domains (4). The number of domains was decided by considering two educational groups (Class 9-12 -group 1 and, Bachelor level and master level-group2) and two sex groups (men and women). This sample size will allow the findings to be stratified into four age-sex estimates.
n = 768*4 = 3072
– This sample size was adjusted for the expected non response to obtain the final sample size. An 80% response rate was assumed for the study. To adjust for non- response, the above sample size was divided by the expected response rate.
n= 3072/0.8 = 3840
– This gave the sample size 3840 individuals from the population in the 15-24 years.
However, because of the budgetary constraint, I just want to include 1500-2000 sample size. How can we reduce the sample above sample size into 1500 to 2000. Please give the ways of reducing the sample size.
Hello Bhakta,
To the extent that educational level and sex do account for differences in either behavior (smoking / drinking), the stratification of your overall sample will reduce the needed N for estimating overall/combined population values, as implied by Kosta Sotiroski 's response.
However, you appear to want estimates of comparable precision across all education/sex combinations, so, unfortunately, that doesn't help your situation. What you may be able to do is to identify additional stratification variables (e.g., family income, geographic locale) which are related to the target behaviors and sample proportionally with respect to those variables. A second point is, have you applied the finite population correction to your sample size estimates? Doing so will lower the target amount a bit.
Stratification or accepting less precision are about your only options if you are constrained to a total sample of about half of what you've determined would be necessary. Of course, modifying the solicitation (perhaps with some no-cost or available incentive) to increase the response rate would help as well.
Good luck with your work.
You can use the optimal sample size for which there are many formulas for how to calculate it. And when you have decided to reduce the number of elements in the sample, then you can do this using a multi-stage sample or a stratified sample with a predefined number of elements in the sample.
Hello Bhakta,
To the extent that educational level and sex do account for differences in either behavior (smoking / drinking), the stratification of your overall sample will reduce the needed N for estimating overall/combined population values, as implied by Kosta Sotiroski 's response.
However, you appear to want estimates of comparable precision across all education/sex combinations, so, unfortunately, that doesn't help your situation. What you may be able to do is to identify additional stratification variables (e.g., family income, geographic locale) which are related to the target behaviors and sample proportionally with respect to those variables. A second point is, have you applied the finite population correction to your sample size estimates? Doing so will lower the target amount a bit.
Stratification or accepting less precision are about your only options if you are constrained to a total sample of about half of what you've determined would be necessary. Of course, modifying the solicitation (perhaps with some no-cost or available incentive) to increase the response rate would help as well.
Good luck with your work.
@ Bhakta KC
You have a few chances:1. you can reduce the level of precision that is a marginal error (>5%).
2. you can reduce your domain 4 into domain 3, then form each domain collect the sample from 50% men and 50% women. 768 in bachelor level so 384 from men and 384 from women or according to their total number of population. then here you will get data as sex group, you have not to collect the data differently.
3.you assume current tobacco user 50%. so, you can reduce that (
Hello Bhakta KC
Let assume sample proportion of tobacco user = 50% = 0.5
Let assume that margin error =5% = 0.05[this should be consider below 5%] = E
confidence level = 5% = alfa
the critical value at 5% level is 1.96 = Zalfa/2
to determine sample size formula is n =[Zalfa/2]^2*p*q/E^2
therefore n = [1.96]^2*0.5*0.5/[0.05]^2 = 384.14~384
this is the approximation sample size value of the cross sectional study done in school/college level students about their smoking and drinking behaviors, may be the sample size is vary if your will consider the margin error below 5%, generally margin error is less we will get more accurate results.
you can use sample size formula based on marign of error, standard deviation and level of significance.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size