Research Tips
Representativeness and Generalisability
In formal research contexts, researchers often strive for representative samples. What this means is that those participating in the research are similar to the whole group to which the results will be applied; in other words, that the results will be generalisable. This is important because if those who participate in research are somehow different from those who do not, predictions and decisions based on their attitudes and behaviours as expressed through the research may not hold up in real life. Even in carefully designed research projects, there is the potential for non representative samples to result from nonresponse bias or, common in less well designed projects, self-selection bias.
Even outside the context of formal research, it pays to think about representativeness and generalisability. For example, managers have long made decisions based on particular customers they’ve talked with or based on their own experience with products. And these days, managers can read customer comments in social media such as blogs and tweets. These can provide good initial clues about things to investigate further but should not be considered to generalise to all customers without additional investigation. Managers themselves are very unlikely to be representative of all customers because they are much more interested in and knowledgeable about their own products. Customers who praise or complain about companies — whether in person, on a blog, or via a tweet — are also unlikely to be representative. A relatively small proportion of customers tend to express themselves in these ways and they tend to be those who either have a particularly strong (positive or negative) opinion about the product in question or to be particularly enthusiastic about expressing themselves in general. Sometimes, attempts to generalise from information gathered in these less formal ways can also be threatened by availability bias.