When the sample is not representative, it is said to be "biased." There are two types of bias, according to Wimmer and Dominick (1996):

1. Non-response bias: only a part of the population enters the sample, which is self-elected.

For example: The company's sales manager wants to know the level of satisfaction of his customers with the computers purchased. It is decided to send a questionnaire to all last year's buyers asking them to return the answer by mail with their opinion on their computer. You receive 4% of responses, of which 78% are dissatisfied with your computer. The sample has been selected by the voluntary response of the population and there has been a non-response bias: those who answer the survey are the most dissatisfied with their purchase and those who have no problems often ignore the survey. If there is a very high rate of non-response, it is necessary to investigate the causes, because people who do not respond may have a very different opinion than those who respond.

2. Selection bias: only selected elements with some “arbitrary” criteria enter the sample

For example: To find out the Internet interest of the inhabitants of a city, a survey is sent by e-mail. The way the sample is chosen will lead to answers in favor of the interest on the Internet and does not represent the entire population: there has been a selection bias.

Needless to say, any bias should be scrupulously avoided.