If one has a national sample constructed of 5 to 15 year olds, is it then possible to generate statisically significant analyses of sub-sets of this age grouping , say 5 to 10 year olds and 11 to 15 year olds?
In general, you can pull two domains, one for each subclass of age.
However, the statistical significance of these two domains depends, obviously, from the population from which you extract the sample, the sampling system chosen and the method of allocation (if it has been considered).
Therefore, it is essential to know the confidence interval used for sample size and the maximum stochastic error.
From this you can calculate the significance of the two domains.
This, in brief. Good job.
Hi. Yes, you can make the analysis of those subsets. You must also trust that the wide sample is representative of the nation, for a well defined short lapse of time of data gathering, perhaps smaller than one year. Your conclusions would be valid for that period of time. Ok, best wishes.
You can do this, but this would not be very sensible to do.
You should rather model your response variable using age as one continuous explanatory variable (predictor).
And please be aware that "statistical significance" (say, the p-value) is not interpretable when you adjust your analysis "after the fact" (after having the data). I know that this is commonly done, but it is largely nonsense. As I see it you should simply forget about "statistical significance" at all and better provide a sound description and analysis of the data, focussing on visualization and summarization of relevant features. You may use the data to invent/imagine a model that is as simple as possible but sufficiently flexible to model these features, and you can discuss the estimated coefficients in the model (that ideally would have some real-world meaning) together with the (un-)certainty of these estimates. That's all not about "statistical significance". It is rather an attempt to learn from your data (what is much much better than relying on statistical significance).
Thanks to all of you for your replies. Perhaps I confused you by incorrectly using the term 'statistical signficance' when actually I meant 'statistical representativeness'. The data analysis that would be presented would in fact be simple frequency distributions and not associations. But yes the confidence intervals used are critical. Thanks again!
- Badges
- Science topic
- Similar topics
- Medicine
- Epidemiology
- Age Groups