I am working on a study of sample size 100 collected through snow ball technique using referral method. Can you please guide that in order to achieve my objectives, should I apply parametric techniques ( as sample size is 100) or should apply non- parametric techniques ( as samples are not selected randomly) ??
Please pour your valuable suggestions
Kanika -
The difference between parametric and nonparametric techniques is not a matter of randomization, but rather a difference in whether you need to rely upon more detailed information about data distributions, involving shape of the distributions, and perhaps means, or less detailed such information, perhaps rankings of data by size, and median values.
Sample selection is done randomly as an effort to hopefully obtain a sample which is "representative" of the overall population in its 'shape' (which entails moments: mean, variance, skewness, kurtosis). So, yes this is very important to parametric applications. However, it is also important that you do not, for example, draw all of your data from one part of a population, and then use nonparametric methods on that sample, expecting it to represent the population. Of course random sampling does not guarantee that your sample will resemble a small version of your population, but methodologies to estimate variance have been developed, though many do not realize that an unfortunately drawn random sample can give you a very inaccurate estimate of variance as well. (One must keep in mind that the variance in estimating a mean from a sample is not the same thing as the variance of the population, and that in regression, it is the variance of the estimated random factors of the estimated residuals which matters, and not even the shape of the unconditional y population data.)
If your sample is not 'representative' of your population, this will adversely impact results from any method, parametric or nonparametric.
There is literature on snowball sampling, I think, regarding what it takes to expect to be something close to representative. If I remember correctly, it takes about twice as many observations as in random sampling, but I expect that varies greatly. I know very little about snowball sampling, but for both randomized selection sampling and model-based (prediction-based) sampling, stratification is important to help ensure collection of data from across the population, as the population may be basically a mix of subpopulations with different characteristics. I expect that you want to be sure that snowball sampling has done the same thing for you.
But I don't think there is the relationship between randomized versus snowball sampling, on one hand, and parametric versus nonparametric techniques, on the other hand, that it seems that you may think that there may be.
Perhaps I misunderstood your question. Hopefully I'm not looking at this the wrong way. I hope that there is something useful in the above for you.
Cheers - Jim
Thanks a lot, Sir for this detailed explanation of my doubt James R Knaub
Hi Kanika,
It would have been important to learn from your question what is the purpose of your study and weather your project deals with empirical research, or its an evaluation study. Also, how many samples of 100 subjects are you going to employ? If you could revise/rewrite your question and be more specific, I think you would help readers much better to answer it. Best of luck with your project.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size