Sample from a sample

#1
Hello all,

Also posted in Sample of a sample - Cross Validated (stackexchange.com)

If we were to carry out a survey on say, workers in a chemical industry asking for responses such as:
  • Are you a smoker? Y/N
  • Is your BMI over 25? Y/N
  • Do you have asthma?
  • Did you diagnose asthma whilst in the chemical industry?
  • Do you have cancer?
  • Did you diagnose cancer whilst in the chemical industry?
Could we then focus on say a group of the smokers from those who responded and carry out eg blood tests on them - would we be able to calculate any statistical significances. The reason I ask is that it seems that we are taking a sample from an existing sample?
Is there a recommended sample size that we would need?

Thank you.
 
Last edited:

Dason

Ambassador to the humans
#2
It mainly depends on how you're performing the initial survey in the first place. There is definitely not enough information that you've given to provide any guidance on sample sizes.
 

hlsmith

Less is more. Stay pure. Stay poor.
#3
The ideal is for a random sample of employees and everyone participates or non-response is not related to a systematic reason. Given this and an adequate sample size, you can likely generalize the results to the super population under investigations.

However, if you don't have the about scenario, results may not generalized to the population without additional processes. You need to describe your data in more detail, including sample sizes. In addition, you can have the above scenario, but people do not respond honestly, given the face validity of questions, e.g., do you smoke, well yeah but I don't what to tell you that. So, there can be issues stemming from many sources.
 

noetsi

No cake for spunky
#4
Regardless of the size of the sample socially desirable answers will ruin your sample. If people say what they think you want to hear, you could have a million in the sample and it won't matter.

Which is what hlsmith says. I run into this issue when we do our climate survey. Everyone says they are satisfied and then in the comment section they give strong reason to doubt they are satisfied :p
 
#6
To add to noetsi's point, another form of corruption can occur if the question can be perceived has having some impact on the interviewee's life, such as if it was asked during a housing intake process, or in relation to a financial arrangement, or have legal consequences.
 
Last edited:
#8
Hello all,

Also posted in Sample of a sample - Cross Validated (stackexchange.com)

If we were to carry out a survey on say, workers in a chemical industry asking for responses such as:
  • Are you a smoker? Y/N
  • Is your BMI over 25? Y/N
  • Do you have asthma?
  • Did you diagnose asthma whilst in the chemical industry?
  • Do you have cancer?
  • Did you diagnose cancer whilst in the chemical industry?
Could we then focus on say a group of the smokers from those who responded and carry out eg blood tests on them - would we be able to calculate any statistical significances. The reason I ask is that it seems that we are taking a sample from an existing sample?
Is there a recommended sample size that we would need?

Thank you.
If you will consider these questions, I believe your responses will be biased because the questions are biased.
 

fed2

Active Member
#9
Sample, from a sample, from a sample,
Creeps in this petty pace from day to day
To the last syllable of recorded time;
And all our yesterdays have lighted fools
The way to dusty death.
 
#10
Hello all,

Also posted in Sample of a sample - Cross Validated (stackexchange.com)

If we were to carry out a survey on say, workers in a chemical industry asking for responses such as:
  • Are you a smoker? Y/N
  • Is your BMI over 25? Y/N
  • Do you have asthma?
  • Did you diagnose asthma whilst in the chemical industry?
  • Do you have cancer?
  • Did you diagnose cancer whilst in the chemical industry?
Could we then focus on say a group of the smokers from those who responded and carry out eg blood tests on them - would we be able to calculate any statistical significances. The reason I ask is that it seems that we are taking a sample from an existing sample?
Is there a recommended sample size that we would need?

Thank you.
If you will consider these questions, I believe your responses will be biased because the questions are biased.