Let be Statistan a country of approximately 20 million farmers. I want to draw a representative sample of the farmer population in order to survey them, in particular with the goal of capturing their yield (as defined by kilograms per cultivated hectare) and understanding what makes yield varying across the country.

Statistan is a very diverse country.

  1. Its cultivated land is 50% rainfed and 50% irrigated;
  2. Environmental conditions change a lot through the country with 33% of the cultivated land in mountainous regions (very steep fields), 33% in tropical regions and 33% in arid regions;
  3. Social conditions also differ with 50% of the farmers living in regions with of high per capita income and 50% of the farmers living in region of low per capita income.

Should I take into consideration these categories when drawing the sample? Of course these are all important factors in determining each farmer's yield. It seems that if I sub-group my total population before sampling it I dramatically reduce the overall representativeness of the sample. But also if I don't I won't have many observations from less populated areas (like arid regions) where the potential of relevant findings is higher.

How can I better understand the trade-off between sub-grouping and representativeness of the sample population given the number of observation I will collect with the survey?

  • 1
    $\begingroup$ If you want to understand what makes yield vary, wouldn't you want to sample from as many regions with different levels of potentially causal/explanatory factors as you can? If so, it seems to me like the tradeoff is between representativeness and the cost of sampling sufficiently from all such regions. Cost depends on other factors like data collection difficulty and statistical power needs, in addition to the kinds of factors you mentioned that determine the number of regions. Is there some other reason besides cost (time, energy, money, risk) to exclude regions? (To "sub-group," right?) $\endgroup$ – Nick Stauner Jan 5 '14 at 23:49
  • 1
    $\begingroup$ This sounds like an exercise rather than a real study. For a real study we'd want to know how you expect identify farms for selection. As there is rarely a list, some kind of area sample would be called for. As Nick states, the sample allocation would depend on cost information for setting up a survey in different parts of the country. $\endgroup$ – Steve Samuels Jan 7 '14 at 3:58

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.