For both B2B and consumer market researchers the question that lies in the back of our minds is to what degree is our sample a reasonable measure of the population(s) we are interested in? All aspects of our projects can be spot on, but if the respondents do not form a representative sample then the question of data quality arises.
How do we know if our sample is representative? There are many methods available to the researcher to ensure that the sample is free of bias. The first order of business is to make sure samples are selected randomly. There are numerous sources of error that are outside the researcher’s control. Randomization will minimize their impact. In theory, randomization spreads potential bias evenly across all respondents sampled. If you are drawing a representative sample from a known list then there are several methods to create a random sample including simple random selection, stratified sampling or cluster sampling.
If you are conducting tracking studies such as brand awareness or customer satisfaction then timing is important to the process. These types of studies typically draw from a potential respondent pool that is created on an even time basis, e.g. weekly, monthly or quarterly. Samples drawn from different time frames can produce misleading results, due largely to factors outside of your direct control. This can lead to a situation where you end up with something less than a representative sample.
Having benchmark statistics to compare to is critical. This can be easily performed during the survey data analysis phase. If you have a known geographic distribution of sales then the distribution of survey respondents can be compared to this standard. If deviations are noted, for example, one region responded more frequently than expected, then your survey response pool can be weighted after the fact. This is another good reason to invest in a CRM system that captures both transactional and consumer data.
The success of any consumer insight project requires that the samples selected, and subsequent response files align with known parameters. These parameters can be specific to the company or benchmarks set by projects such as the US Census. If making projections outside of your sample is important then you must pay attention to the quality of the sample and the data it provides.