when we are talking about sample then it is advisable to carryout analysis with sample only for result and interpretations.

Thanks for your response. So, you recommend using smaller sample? The original 35000 records were also sample.

Sample size is so high-then what is popn. size?

Sample size calculation

There are several methods used to calculate the sample size depending on the type of data or study design.

Generally, the sample size for any study depends on the:[1]

• Acceptable level of significance
• Power of the study
• Expected effect size
• Underlying event rate in the population
• Standard deviation in the population.

Suppose that I make a calibration curve with the form Y = mx + b. Here, 'x' is the area under the curve as measured by the software. Y is the concentration, say 0.01 to 2.0. If I have 2.0 = (m)*123456789 + 0.1, 'm' needs to be small. To say 'm' is small therefore it doesn't matter means no one has ever quantified the amount of chemical in a sample.

If you want to see how reproducible your results are, break your data into 3 different subsets of data. Create a model with one subset of data. Use that model to see how well it predicts the data in the other subsets of data. After you see how well that works, break your original data into 3 more (and different) groups and repeat your analysis. Repeat those steps a few more times. If the model you get is data dependent, meaning totally different every time, most likely your data is random. If you get close to the same model each time, it looks like your model might be good.

Thanks so much for your comments.

Hello Hanieh,

Yes, statistical significance is much easier to obtain with huge sample sizes.

However, making sense of the results requires thoughtful interpretation of the magnitude of effect: is the difference sufficiently large to be meaningful to decision-makers, practitioners, other researchers, or for influencing theory? That, it seems to me, is your challenge here. For this, you could consult: (a) domain experts; (b) existing research on the same/similar variables; (c) cost-benefit analysis.

I wouldn't ignore the larger sample results (N = 35K) in favor of the "smaller" sample results (N =5K); if anything, you should have better precision of parameter estimates from the larger sample than the smaller sample.

Good luck with your work!

Interpreting statistical significance is meaningless if you don't interpret your results also in terms of their practical relevance. And your test is very likely to be over-powered...

Thanks so much everyone for your help :)

The large sample size will improve the precision in estimating the coefficients. The size of the coefficients is not relevant. The proportion of the variability in the data explained by that variable is important. "This" variable explains 35% of the variability while "that" variable explains 0.005% of the variability in the response variable. While both were highly significant (p

.

as already said above, statistical significance does not imply pragmatic relevance !

regarding effect size measurement, the following link might be useful

Article Understanding the effect size and its measures

.