Size matters in drawing samples. The larger your sample is – up to a point – the smaller the range of variability there is in your results (margin of error). After that point, additional persons in your sample do not add significantly to the accuracy of your results except if you are wishing to look specifically at certain groups.

So you do not need to have samples of 1,000 persons in a survey to make it fairly accurate (no sample survey is 100% accurate). The size of the sample can also be influenced by the amount of time you have to do the study and the amount of money you wish to spend. But you really ought to use the margin of error that you are comfortable with to decide how big a sample you need. In theory, the samples used in both polls appear to be adequate. But we cannot be sure since they do not disclose the details of the samples neither do they indicate whether or not there was weighting of the sample to ensure that they were representative of the population.

The most important thing in examining a study is how the sample was drawn: is it being done randomly or not? The goal in probability sampling is to give each person in the country a known chance to be selected. If you are not using a probability sample then you will not be able to determine the margin of error for the study. (Even though some people quote one, they are fooling themselves).

HHB does not disclose the details of how they got their random sample. However, it is possible to have a random sample from a panel maintained by a pollster. One has to be careful though that the basis of participation was not the choice of the panel member but the choice of the pollster. The issue of refusals to participate affects the sampling process but it can be managed with care. However, if there is not sufficient care taken this route can end up being a non-probability sample. HHB’s track record suggests that they would have taken the requisite care.

SBS reported using the “random digit dialling method”. This method fits the definition of random sampling in that it ensures that all the telephone numbers in the country have a known chance of being selected. It depends on the use of area codes which give the proportion similar to how your residential address is in a fixed location do so that each area code can be given a set number of calls. In this way -638 (San Juan) and 637 (Diego Martin) for example- will have a set number of persons to be called, given the population in their areas and the landline penetration there.  Who actually gets called depends on the random selection of the last two or in some instances four digits. Usually, these are selected automatically and are not left to the interviewer.

Cellphone sampling is more tricky since the numbers are assigned based on where you bought the phone not based on where you live. There is great difficulty in ensuring the location reported by a respondent who uses a cellphone. One needs to use other sources of information to validate the addresses of the respondent. With the coming of certain providers, the numbers for the phone area code may represent either a cell number or a landline. Response rates for cellphone surveys are usually lower than those for landlines and as such the costs are higher. Then there is the problem of combining the samples.