I am conducting a survey in Hayatabad Peshawar ,a modern town ship comprised of 7 Phases and each phase consist of 4 sectors and almost 12000 house in total ,i want to include all of them in the survey ,what sampling technique should i use
you are saying "I want to include all of them in the survey", when the total population is included in the study, then it is not a sample, because a sample means "part of"
If you will take a sample, then the best approach is the stratified random sampling technique is going to give you a good result.
you are saying "I want to include all of them in the survey", when the total population is included in the study, then it is not a sample, because a sample means "part of"
If you will take a sample, then the best approach is the stratified random sampling technique is going to give you a good result.
U suggest using a purposeful sampling strategy, sampling a specific number of households from each sector & phase.
For a general overview on spatial problems (including sampling), see the HSR special issue on spatial analysis.
Book Spatial Analysis in the Social Sciences and Humanities
• A multistage simple random sampling method will be used. Hayatabad is having 7 phases but we will include only first 4 phases in our study because phase5, 6 and 7 are new phase and population there is not stable. Each phase consists of several sectors and each sectors having several streets. In first stage phases will be selected randomly and then sectors in each phase will be selected randomly and then streets from each sector will be selected randomly and then houses in each street will be selected randomly and from each selected house data of all those members will be taken which fulfill our inclusion criteria to collect a sample size of 422 participants.
this was the initial plan,now i want to include all of the 7 phase .
Iftikhar -
Is a "phase" in your survey a separate geographical or otherwise separate grouping, with no overlap with another "phase?" Are you doing inferences for each phase separately? And are you doing your multistage random sampling for each 'phase?' If so, and if it seems that your design worked well for each of the first four groups/phases, then is there some reason not to use the same design again, in the case of groups 5, 6, and 7? If you can imply anything about sigmas from the first four phases, that relate to phases 5, 6, and 7, then that may help in determining reasonable sample sizes to be used in those latter phases.
If you want to compare the efficiency of different designs, by the way, you might consider the "deft" noted, for example, here:
http://www.jos.nu/Articles/abstract.asp?article=11155,
Journal of Official Statistics, Vol.11, No.1, 1995. pp. 55–77, Methods for Design Effects, by Leslie Kish.
You might then compare your design to a simple random sample, to see how helpful it may seem to be. However, to speculate if another design would be better is more problematic. It would seem you would have to make some somewhat subjective assessments.
Would cluster sampling be of use? I am not sure of your application. Note that cluster sampling actually uses data less efficiently than simple random sampling, whereas stratified random sampling is designed to be more efficient. But if traveling is a problem, cluster sampling can be more cost effective.
Regardless, please consider "total survey error" (TSE): sampling error and nonsampling error, variance and bias. Starting with variance, designed to estimate sampling error, but actually impacted also by nonsampling error, and considering the finite population correction factor, if substantial and you are addressing inference from a finite population, you always want to assess accuracy of your results in the end, and sample size needs and design considerations, at the outset.
It is good that you are using random sampling, as that is useful for making statistical inferences. Also, if you have regressor data, say on groupings of continuous data for which a model applies to each grouping, not necessarily the same model in each case, then you may produce good inferences, but I doubt you have good regressor data on the entire population. If you do, then combining such 'predictive' modeling with random sampling may be very useful for you. That falls under the category of model-assisted design-based sampling and estimation. My expertise is largely with continuous data, but there are generally analogous considerations with proportions (yes/no data).
Best wishes for a successful and useful sample survey.
Cheers - Jim
PS - I just noticed that it sounds like you may be talking about "area sampling." You may want to do an Internet search for articles and other information on that topic.
l think stratified random sampling is the best. Read more on stratified
- Badges
- Science method
- Similar topics
- Mathematical Sciences
- Statistics
- Sampling Design