Non-probabilistic sampling is suitable for shopping malls. They sampled at random to get a higher quality report.
Probabilistic sampling is also called random sampling. Probabilistic sampling is based on probability theory and random principle, so that each unit in the group has a pre-known non-zero probability. The probability of the whole unit being selected can be specified by sample design and realized by some random operation. Although random samples are generally not completely consistent with the whole, the sampling error can be calculated and controlled based on the law of large numbers, so it can correctly explain to what extent the statistical value of the sample is suitable for the whole, infer the whole quantitatively according to the results of sample investigation, and explain the nature of the whole to some extent. Features. Probability sampling is mainly divided into simple random sampling, systematic sampling, classified sampling, cluster sampling and multi-stage sampling. In real life, most sampling surveys adopt the method of probability sampling to extract samples.
principle
The basic principle of probabilistic sampling is that the larger the sample size, the smaller the sampling error, while the larger the sample size, the higher the cost. According to the law of mathematical statistics, when the sample size increases linearly (double the sample size and double the cost), the sampling error is only the square root decrease of the relative growth rate of the sample size. Therefore, the design of sample size is not as big as possible, and it is usually limited by economic conditions.
principle
Market research methods
A desk research case study method B non-repeated sampling C sampling survey reset sampling lottery method product lien test D multidimensional scale quantitative research legal research method Typical survey method Telephone survey Multi-stage sampling equidistant sampling Independent control quota sampling isometric scale E Second-hand data survey Two-way focus group F non-probability sampling stratified sampling stratified proportional sampling stratified optimal sampling G observation, probability sampling, inflection point survey, snowball sampling, H conference survey, J focus interview, Empirical judgment, random sampling, family diary, dealer interview, K feasibility study, L joint analysis, lien survey, garbage survey, category scale, M interview, blind survey, descriptive survey, PP PPS judgment, sampling quota, sampling balance scale, evaluation scale, paired comparison scale, Q Q classification, R arbitrary. Sampling s capacity measurement method SEM model depth interview method double sampling experimental investigation method field survey numerical distribution scale random number table method sequential scale T projection technology extension estimation method projection research exploratory research W literature survey method questionnaire survey method network survey copywriting survey method unprepared visit online survey X inquiry method syndicate survey whereabouts analysis mutual control quota sampling Y mailing survey causal survey Z subjective probability method cluster sampling focus survey door-to-door search method.
Probabilistic sampling can ensure the representativeness of the sample to the whole population. Its principle is that it can form samples according to the probability of various random events contained in the internal structure of the group, making the samples a microcosm of the group.
superiority
(1) probability sampling includes the following advantages:
Investigators can get information about people of different ages and levels; The sampling error can be estimated; The survey results can be used to infer the population. For example, in a survey using the probability sampling method, if 5% of the respondents give specific answers, then the investigators can combine this percentage with the sampling error to summarize the overall situation.
On the other hand, probability sampling also has some shortcomings:
-In most cases, the cost of probabilistic sampling of the same scale is higher than that of non-probabilistic sampling;
-Probabilistic sampling takes more time to plan and implement than non-probabilistic sampling;
-The sampling plan implementation procedures that must be followed will greatly increase the time for data collection.
way
Probability sampling includes simple random sampling, systematic sampling (equidistant sampling), stratified sampling (type sampling), cluster sampling, multi-segment sampling, PPS sampling and indoor sampling.
Equidistant sampling
In quantitative sampling survey, equidistant sampling often replaces simple random sampling. Because this sampling method is simple and practical, it is widely used. The samples obtained by equidistant sampling are almost the same as those obtained by simple random sampling. The basic method of equidistant sampling is to arrange and number the units in the population in a certain order, and then decide an interval, on the basis of which the units and individuals under investigation are selected. Sample distance can be determined by the following formula: sample distance = total unit ∕ sample unit.
For example, suppose you use a local phone book and determine that the sample distance is 100, then 1 samples are taken from 100. This formula ensures the integrity of the whole list.
The equidistant sampling method uses a starting point at will. For example, if you use a phone book as a sampling box, you must take out a number at random and decide to start browsing from that page. Suppose you start from page 5, choose another number on that page and decide to start from this line. Suppose you choose to start from line 3, which determines the actual starting position.
Compared with simple random sampling, the main advantage of equidistant sampling is economy. Isometric sampling is simpler, less time-consuming and cheaper than simple random sampling. The biggest disadvantage of using equidistant sampling method lies in the arrangement of the whole unit. Some general units may contain hidden tables or "unqualified samples", and investigators may neglect to select them as samples.
group sampling
Stratified sampling in quantitative survey is an excellent probability sampling method, which is often used in AIA previous surveys.
The specific procedure of stratified sampling is to divide the overall unit into two or more independent and complete groups (such as men and women), and conduct simple random sampling from two or more groups, and the samples are independent of each other.
The whole units are grouped according to the main signs, and the signs of grouping are related to the overall characteristics we care about. For example, we are doing a survey of beer brand awareness, and preliminarily judge that men's understanding of beer is different from that of women, so gender should be an appropriate rank symbol. If stratified sampling is not carried out in this way, stratified sampling will not get any effect, and no amount of time, energy and material resources will be wasted.
Compared with simple random sampling, stratified sampling is often chosen because of its obvious potential statistical effect. That is to say, if we take two samples from the same population, one is a stratified sample and the other is a simple random sample, then the error of the stratified sample is relatively small. On the other hand, if the goal is to obtain a certain sampling error level, then smaller stratified samples will achieve this goal.
In the investigation practice, there is actually a certain cost in order to improve the accuracy of stratified samples. Usually, our realistic and correct stratified sampling generally has three steps:
First, distinguish prominent (important) demographic characteristics from classification characteristics, which are related to the behavior studied. For example, when studying the consumption rate of a product, it is common sense that the average consumption rate of men and women is different. In order to take gender as a meaningful indicator of stratification, investigators will certainly be able to produce data to prove that the consumption levels of men and women are obviously different. In this way, various features can be identified. The survey shows that, generally speaking, after six important salient features are determined, increasing the distinguishability of salient features is not helpful to improve the representativeness of the samples.
Second, determine the proportion of the whole at all levels (if gender has been identified as a prominent feature, what is the proportion of men and women in the whole? )。 Using this ratio, the number of people in each group (layer) in the sample can be calculated.
Finally, researchers must extract independent simple random samples from each layer.
Nested sampling method
All the above sampling types are extracted by unit, that is, each unit is extracted one by one according to the number of sample units. In cluster sampling, samples are grouped in units.
Cluster sampling has two key steps:
Homogeneous groups are divided into completely smaller and independent subsets.
Randomly select subsets to form samples.
If the investigator observes all the units in the selected subset, we have a first-class cluster sample. If we extract some unit observations from the selected subset in a probabilistic way, we will get a two-level clustering sample. Both stratified sampling and cluster sampling should divide the population into independent complete subsets. The difference between them is that stratified sampling samples are extracted from each subset, while cluster sampling is to extract some subsets.
Geographical area sampling is a typical cluster sampling method. Investigators who go door-to-door to investigate a specific city may randomly select some areas and visit some groups more intensively, which greatly reduces the visiting time and funds. Cluster sampling is considered as a probabilistic sampling technique because it randomly selects groups and units. It is worth noting that under cluster sampling, we assume that the units in the cluster are as heterogeneous as the population. If the characteristics of units within a group are very similar, if the differences within a group are small, the differences between groups are large because of the same environment. Generally speaking, to solve this problem, we can expand the number of groups and then extract a small number of units from each group to ensure the representativeness of the samples.