Under ideal conditions, a sample should be the exact representation (of smaller size) of the population that you wish to probe. Obviously, it is rare that reality is ideal. Also, all methods have disadvantages. Then, it will be important to make a careful choice.
When you want to build a sample, your goal should be to avoid any potential gap between the population you're probing and your sample. If so, you can consider the sample to be "representative". The key to this objective, it is to find the method whose disadvantages are most controllable. With this intention, there are various possible strategies. One can gather them in two families
- The family of the probabilistic strategies.
- The family of the non-probabilistic strategies.
The probabilistic strategies are the most conventional. They are based on the law of the great numbers. In gross words, this law states this: if you "pick" a great number of individual cases, randomly, the global characteristics of that group of individual cases will be in the same proportions as in the original population. It is for this reason that it is not necessary to carry out 7 million interviews to make a pre-election survey in Quebec (Canada). In fact, a few thousands are sufficient to obtain a precision more than adequate. In addition to the quantity of respondents, the level of accuracy is related to the number of possible choices. Thus, if 2 choices are possible, you will need less cases than if there are 3 or 4 choices of answers. If the number of the choices increases, it is necessary to probe a greater number of people to obtain same reliability. With an infinite number of cases you have the "normal curve" used by statisticians. This curve is the subject of the article "the normal curve used in statistics"
The non-probabilistic strategies are more hazardous. There is no rule to quantify the precision of information that are gathered by these methods. For this reason, several look at these techniques with skepticism. The precision will depend solely on the care which the sounder will take to avoid the risks of any error. It is vital, for those who use these methods, to thoroughly describe the methodology used. Moreover, the sounder must be able to demonstrate that he took care of any source of a potential problem. Under these conditions, those who will use these results will be able "assess" any risk of imprecision even if it is impossible to quantify it.
For those which use the results of surveys carried out using non-probabilistic strategies never forget that you will never be able to quantify the level of confidence of your survey. Some, like to give an evaluation based on the number of respondents in order to indicate the precision of their survey. Such a figure is completely erroneous for non-probabilistic strategies. The formulas making it possible to calculate it are based on the "laws of the great numbers". Such figures are thus valid only when the respondents are selected in a truly random way.