I studied many years ago, that an unbiased survey based on 25 replies give quite reliable results.
I now have the chance of seeing this theory put into action. I have analyzed the survey results based on2 5 replies, and compare the results with 50 and 100 replies. The results given are quite close. It means that the results based on the first 25 replies are quite reliable.
I shall be showing the results for 25, 50 and 100 replies to various surveys here, if the survey did reach 100 replies.
Quite true.
ReplyDeleteThe sampling theory says that if the sample size is at least 30, we can use statistical indicators to represent the whole population.
But Singapore is very small and hence also very homogeneous! Just like one GRC is not much different in population mentality from another hence if PAP can win in one with over 60%, same for other GRCs and this has also been proven in every election. That's why PAP can win big every time because of this. You think the same can happen in Malaysia for the urban and rural areas?
ReplyDeleteWhether it is biased or not would dependent whether these 25 are really representative.
ReplyDeleteThe problem lies in that it is web-based and surveyed openly. And you could not select the "population" to be representive unless you send questionaire by email to the selected "population".
In a conventional survey, you would have to sent questionaires to about 100 persons and the respondents must be more than 20% to be representative of the population surveyed. Hence, 25 respondents is a good sampling, if your questionaires reached the 100 persons selected.
And the population of 100 surveyed must also be representative. For instance, if you are conducting a survey on property investments, school children could not make up this 100 persons.
Unless questionaire is sent by email to selected targetted population individually, the results could still be biased.
To "December 27, 2008 12:18 PM",
ReplyDeleteThe 30 sampling that you are referring to is the central limit theorem which states of is if the samples are taken randomly and independently than the expectation of the sample is an unbiased estimate of the population. This unbiased estimator has a normal distribution of the mean equal to that of the population and a standard deviation of the standard deviation/squartroot(sample size). If the standard deviation of the population is unknown, we can approximate the sample mean to that of a Student's T distribution with Sample Size - 1 degree of freedom of which the standard deviation is the sample standard deviation divided by sampling size.
From Mr. Tan's short description, the 25 replies is less than 30. Also it does not appear to be randomly selected because it requires people to key in their feedback on the website. The selection of the samples are obviously not random. Thus, the 25 replies cannot be relied on.
Ah so.
ReplyDeleteAlthough not trained in stats, I remembered reading that actuaries and statisticans, were vv careful abt using vv small samples.
Knowing yr background, I was wondering how you assessed small samples.
Well we now know.
If the target group is, say #,000 or even ##,000, and a random sampling of the opinions of the first 25 is picked by, say drawing lots, then the aggregate or average opinions of this group of 25 may be more likely than not to be representative of the entire group.
ReplyDeleteHowever, if relying on specific medium for response, e.g. TV, SMS, newspaper or internet, the random quality of respondents would be compromised by the medium used. For example, 25 internet users cannot adequately represent a target group of 1,000 as a random sampling of 25 if only 500 of this group are internet users.
PAP MP Ms. Denise Phua once noted before GE 2006, 85% of forummers and bloggers seem to be anti-PAP. However, the results of the GE showed a big discrepancy as only less than 35% actually voted for the PAP.