How to solve selection bias in a survey? Selection bias often happens in survey studies. For example, sending out a survey to 10,000 customers to ask whether they like one specific product or not. Only 10% of people respond and 80% of them say like the product. How about the other 90% of people who do not answer the survey? Is this selection bias a big concern? Is there any way/method to solve this selection bias? 
 A: Yes, selection is a major confound. You can think of the problem like any model dealing with endogeneity. The predictor is the decision to respond to the survey, and the outcome variable is the rating of the product. The relation is confounded because both may be dependent on attitude toward the product.
You could solve the problem if you had an instrument--a variable that predicted decision to respond but not rating of the product. Random selection of respondents serves that purpose, but the selection confound creeps back in if respondents don't feel compelled to respond even though they have been randomly chosen.
So maybe you have another instrument? If you have a history of conducting these surveys, you might have identified demographic variables that historically predict decision to respond but not rating of the product. Those variables might serve. And there are other statistical techniques designed to eliminate confounds without the need for instruments (like copulas) which are fairly sophisticated.
A: It's a huge concern in all sorts of areas. 
For the example you give (liking a product) people are more likely to answer if they either really like a product or really hate it. 
More generally, people in different demographic groups are more or less likely to answer surveys -- and it also depends on how the survey is given (internet, phone, in person, postal mail etc) and what the survey is about (products, politics, behavior).
There is a large literature on how to increase response rates to surveys; I'm not up on the latest (I last looked at this in grad school, which was >20 years ago) but Googling should turn up literature. 
However, I do know that, for many surveys of products, the goal is not to get an overall rating of the product but to identify its best and worst points. 
