همه چیز درباره RESTful API (رست فول)
February 28, 2023
Full introduction of 4 deep learning algorithms
April 19, 2023
Before inferential statistics, we described descriptive statistics in a separate article and explained some of the methods used. In this article, we are going to provide explanations about inferential statistics.
Let’s start with the question, “If you are going to examine the behavior of the people of a large country, can you test each individual in that country and look at the conclusion?” Obviously, even if this is possible, it is not affordable because it requires a lot of time and money. Here, the main point in the discussion is inferential statistics, which includes more advanced methods than descriptive statistics.
What is inferential statistics?
Sometimes we do not have access to the community in statistical analysis, but we need to be able to analyze it and get some useful information. In this case, the simplest way is to get a smaller set from the society in question and then analyze it. In fact, we have to sample a statistical community. Finally, we must be able to generalize the information obtained from the analysis to the larger community. These are actually possible using inferential statistics.
Since we do not have access to all statistical population in inferential statistics, we must examine a small sample of the population and generalize it to the larger population.
Inferential statistics versus descriptive statistics
As we mentioned earlier, in inferential statistics, we should examine a smaller set of the statistical population and generalize the result to the statistical population. In fact, the statistical population is the population on which the surveys are supposed to be conducted. Sampling in the statistical population is done in different ways, some of which are: random, stratified and quota sampling.
There are more methods that are discussed below. In general, in statistics and probability, sampling methods are divided into three categories: probable, non-probable, oblique and non-oblique. Choosing a suitable sample from the community is one of the most important steps in inferential statistics.
While in descriptive statistics, we intend to describe a large amount of data with graphs and summary tables, but we do not draw conclusions about the population from which the sample was taken.
In descriptive statistics, symbols are Latin and in inferential statistics, symbols are Greek.
Sampling methods
There are various methods in statistics and probability for sampling from the statistical population. Each of these methods is in a separate collection and their implementation is different. The most important issue in inferential statistics is sampling. As we mentioned earlier, we have three categories for sampling:
1- Probability
- random
- clustered
- Systematic
- Stratified
2- Non-probability
- judgmental
- snowball
- Availability
3- Oblique and non-oblique
In the following article, we introduce various types of inferential statistics.
Nonparametric inferential statistics
Nonparametric inferential statistics are used when we have little information about the distribution of the data. In fact, we perform our statistical tests with fewer assumptions. Depending on his or her goal in non-parametric inferential statistics, the researcher can use different methods, some of which include variance analysis, covariance analysis, correlation, etc.
On the Bigpro1 website, you can easily analyze your datasets and extract useful information from them. Just go to the data mining section in your user profile and then click on the statistical analysis option. In the next step, it is enough to select one of your datasets and choose one of the statistical analysis methods depending on your purpose. In bigpro1, it is possible to use non-parametric tests.
Parametric inferential statistics
When using parametric inferential statistics, we are aware of the condition of the data in the statistical population, and these tests are mostly used for quantitative data, which usually have a normal distribution. When we have information about the shape of the data distribution, parametric correlations are more preferable than parametric tests.
Accuracy test
In the accuracy test, you can evaluate the accuracy of your model when using binary data (data that has two states) for correct values and obtain different measures of the convolution matrix. Even if you don’t have a specific threshold limit, you can still choose one of them by using the roc curve chart and obtaining three values of the upper, lower and middle threshold limits and consider it as the threshold limit.
Correlation test
This test calculates the degree of correlation of features; This test is used in both descriptive and inferential statistics. In Bigpro1, you can determine the type of data (quantitative or qualitative) and select at least two columns to determine the degree of correlation between these two features using statistical hypothesis tests.
Analysis of correlation coefficients
In simple words, the analysis of correlation coefficients obtains coefficients that show the degree of dependence of statistical variables. This coefficient is between one (direct) and negative one (inverse) and is equal to zero if there is no relationship between two variables. The meaning of direct relationship is that as one variable decreases, the other dependent variable also decreases and vice versa; And the meaning of the reverse is that when one decreases, the other increases.
