Re: Applied Statistics

[Tuan V Nguyen <t.nguyen@garvan.unsw.edu.au>]

Hi anh Huy et al,

	I take a rather pragmatic view that all surveys are 
subject to bias. The real issue is the magnitude of bias, 
rather than bias per se. I guess you can compensate this 
problem by increasing the sample size. 

	Now, in the case of 

>>An survey conducted by the National Assembly found that 86% of 5,000
>>randomly picked voters across the country said they would vote for a
>>Communist Party member, an Assembly representative said.

while the sample size is impressive, I wonder whether the 
result is that reliable. You see, the survey was carried 
out by people at the National Assembly, not by an 
independent academic institution or agency, so the result 
was not "blind" enough.


>Although applied statistics uses priciples that are simple 
>theoretically, there are more to the difficulties in applying 
>than to meet the eye.

	Yes. As I said before it is like a bikini, what it 
reveals is interesting but what it conceals is vital!

>Take for example a survey of extra-marrital affairs. It is 
>claimed that 10,000 married persons are randomly picked, and 
>ask if they are having an affair and only 1% say yes.
>
>First of all, we have to ask how random the selection of the
>sample really is. For example, the surveyor might walk down 
>the street and pick houses out at random and if there is a 
>married person in the house then you interview them. This might
>seem random enough, but it isn't. Younger people are less likely
>to be at home and they might be the ones who are more likely to have 
>affairs. Couples in which both the husband and wife work away from 
>home are less likely to be at home and they might be the ones who
>are more likely to have affairs. Also, you might pick a house and
>say to the occupant that you're doing a survey on affairs and he 
>says that he's busy, so you pick another house in your quest to 
>interview 10,000 people. If a person is likely to refuse because
>he is infact having an affair, then you sample will be biassed. 

	Oh dear, sao ma` anh ra`nh ta^m ly' cu?a ma^'y 
ngu+o+`i  na`y qua' chu+`ng va^.y? :-))


>Secondly, we have to be careful of interview bias. Eg, if you are
>ask if you are having an affair and you spouse or children is present
>then if you are you would be too worried about the consequences to
>say `yes'. Or you are standing for National Assembly election and
>you are worried that the answer `yes' might be used to disqualify or
>blackmail you.

	This is called "white coat" effect. Hypertension is 
normally increased when it is measured with the presence of 
a doctor. On the other hand, when the measurement is 
performed by a nurse, then it is OK!

	This is also a logistic problem. In quality of life 
study, if you let patients to fill the questionnaire 
themselves then they tend to do it more reliable than you 
ask them. 


>Thirdly, if some of the 10,000 are so diplomatic that you can't
>work out whether their answer is yes or no, then you will have to
>worry whether this introduces bias into your result.
>
>Huy 

	This is particularly true in Vietnam. Saying YES/NO 
does not mean YES/NO per se, but may be a polite way of 
getting away the question! 

	I remember when I first came to Australia, I worked 
in a factory with a VNese friend. This man chose to answer 
his boss' questions based on his boss' facial expression: 
when the boss was not friendly, his answer was NO; 
otherwise when the boss was happy, his answer was YES. It 
worked 80% of the time! :)

	Tuan