Quota sampling is a method used in selecting a sample of individuals to be surveyed, and has been used in most of the local adult Health and Wellbeing Surveys.
Quota sampling involves producing a quota or target number for each group of individuals to be included in the final survey. The quota is generated prior to the survey fieldwork. Quota sampling means achieving a sample of survey responders that are representative of the total population is more likely. As health status and most lifestyle behaviour, such as smoking, alcohol consumption, physical activity, etc, differ by age and gender, and by geography / deprivation too, it is important that there is a good spread of survey responders by age and geography, and between the genders. Thus the groups chosen for quota sampling locally have generally been gender, age group and some geographical component such as ward.
In all surveys, the aim is to survey a group of individuals in order to make inferences about the overall population in relation to one or more measures such as estimating the current smoking prevalence, and this can only be completed successfully if the selected survey responders are representative of the overall population. So if quota sampling is used it can improve the quality of the resulting survey, but it is not always possible to complete the survey in this way, and can be more expensive to use this method of selection, and can only be used in some surveys for practical reasons.
The Office for National Statistics produces annual resident population estimates by gender and by single year of age for different geographical areas including electoral wards for each local authority. These population estimates were used to form a quota for the local surveys. Most of the local surveys aimed to achieve a sample size of 4,000 adults aged 16+ years. Hull’s population aged 16+ years from ONS is approximately 207,000, and thus a sample of 4,000 adults would mean sampling 1.9% of Hull’s population. By using quota sampling, the local surveys aimed to survey 1.9% of all different groups of individuals based on gender, age group and ward. For instance, ONS estimate in 2020 that there are 271 men aged 16-19 years who live in Avenue ward and that there are 551 women aged 75+ years who live in West Carr ward. Using quota sampling with a target sample size of 4,000 residents would involve sampling in equal proportions to the overall population, and thus would involve aiming to achieve a total of five men aged 16-19 years who live in Avenue ward (1.9% of 271) and 11 women aged 75+ years who live in West Carr ward (1.9% of 551).
The method cannot be used in all surveys as it can be impractical in many cases, such as a postal or online survey. However, in the local surveys, survey responders have – in general – been approached through interviewers knocking on their doors and asking if the person answering the door or a household member would be willing to take part in the survey. Where the household was occupied by more than one person, then the interviewer asked about other people in the household if they had already completed their ‘quota’ for the person who answered the door.
Some of the national surveys apply weightings to the survey results in order to make their sample more representative, but this can mean that if there are a small number of individuals in a specific group (say defined on the basis of gender and age) are not representative of the overall population, then the bias can be magnified for particular groups. For instance, if a national sample aimed to have around 500 survey responders in Hull (giving a sample of around 0.24% of Hull’s population) and ONS estimate that in 2020 there are 5,629 women aged 16-19 years who live in Hull then the national survey might aim to achieve around 14 women aged 16-19 years participating in the survey. If they only achieved eight women, then they might apply a weight of 1.75 (so any results would equate to 14 women participating – 8 x 1.75 = 14). However, 14 out of 5,629 is a relatively small number and eight is even lower, so it becomes more likely that the people participating in the survey are different from the ones not participating or from the overall population. This means that the survey is more likely to be biased and the results do not generalise well to the overall population. Applying weighting is good idea if the overall population (in each group) is representative as it means that overall estimate (say of smoking prevalence) is estimated much more reliably, but if the sample of survey responders is biased and not representative of the overall population, then applying weighting does not improve the overall prevalence, and means that the results could become even more biased.
Also see: Bias.