|
Please refer to last week’s article, Plus ça change – the impressively flawed BA survey by Karl Schembri, concerning the Audience Survey which is carried out by the Broadcasting Authority.
The article gives the impression that the sample size will not be representative as it carries out a high margin of error.
Schembri states that a sample of 30 respondents a day would be much more representative, giving more reliability in the cross-tabulation of factors such as gender, age and timeslots.
Although a larger sample size will narrow the margin of error may I point out a couple of facts relating to this. First of all the results will be representative at national level on a quarterly, bi-annual and annual basis, and not on a daily basis. Secondly, when designing a survey a multitude of factors have to be taken into consideration, including both sampling and non-sampling error. One important feature of this survey which Schembri highlighted is the concept of data being collected continuously, rather than at a point in time. Whilst a researcher can measure sampling error, non-sampling error is difficult to measure and every effort must be made to minimise it. As Schembri highlighted in his article, carrying out the audience survey on an ongoing basis addresses this issue by eliminating the possibility of TV and radio stations trying to get wind when the survey will be carried out. This places all TV and radio stations at equal level, as effectively the stations know when the data will be collected. Whilst the previous design was representative of the particular week the survey was carried out, spreading the sample over time takes into account the influence of political, social, and other events happening over time on the audience's preferences, hence the results will be representative of the period of time during which the survey is being carried out.
Determing a sample size is related not only to error but other factors, for example output, etc. With a net sample size of 500 respondents on a quarterly basis and assuming that any one station is favoured by 50 per cent of the population the margin of error is estimated at around 4.5 per cent, meaning that the true proportion of the target population which prefers this station most can lie anywhere between 45.5 and 54.5 per cent. This is a conservative approach and as one moves towards a situation where a station is preferred most by 100 per cent of the population the margin of error will move towards 0 per cent. This can be shown both theoretically and practically.
Increasing the sample size will decrease the margin of error, however these are not directly proportional. For example if the net sample size on a quarterly basis is 2,500 individuals the margin of error would be around 2 per cent for a station which is favoured by around 50 per cent of the target population.
As I indicated previously, the sample size is also dependent on the output, however at this stage the Broadcasting Authority requests that the results are representative at national level. If cross-tabulations will be produced then the methodology which is adopted will incorporate these in the design, ensuring that different categories for which results will be produced separately, eg. gender, etc, will be adequately represented.
From Herald Bonnici
Manager
Research and Methodology Unit
Director General's Office
National Statistics Office
Valletta
|
