What Is Design of Experiments?
Experimental design refers to a kind of planned research, including a series of intentional changes to process elements and their effect observations, and statistical analysis of these results in order to determine the relationship between process variation, thereby changing the process.
- Chinese name
- experimental design
- In the late 60s, Professor Hua Luogeng advocated and popularized the "optimal methods" in China, such as the golden section method, the fraction method, and the Fibonacci sequence method.
- Mathematical statisticians popularize the "orthogonal design" method in the industrial sector.
- In the mid-1970s, the preferred method achieved remarkable results in all walks of life across the country.
- In 1978, due to the missile design requirements, the Seventh Engine Department proposed a five-factor test, hoping that the number of levels of each factor was more than 10, and the total number of tests did not exceed 50. Obviously, neither the optimization method nor the orthogonal design could be used. Subsequently, Professor Fang Kaitai (Institute of Applied Mathematics, Chinese Academy of Sciences) and Academician Wang Yuan proposed the "uniform design" method, which has achieved results in missile design.
- Optimal experimental design The status and significance of experimental design in scientific research:
- 1. The experimental design method is a general technology and a technical method that must be mastered by contemporary science and technology and engineering technicians.
- 2. Arrange experiments in a scientific way to achieve more and better production and scientific research results in the shortest time with minimum human and material consumption. Referred to as: multiple, fast, good, and provincial.
- Optimized experimental design Experimental design can be applied to:
- Improve test efficiency, optimize product design, improve process technology, and strengthen quality management. Experimental design can play an important role in industrial production and engineering design and scientific research, such as: Increase output Reduce quality fluctuations and improve product quality levels Shorten test cycles for new products Reduce costs Extend product life , Electronics, materials, construction engineering, building materials, petroleum, metallurgy, machinery, transportation, electricity [1]
- Experimental design refers to a kind of planned research, including a series of intentional changes to process elements and their effect observations, and statistical analysis of these results in order to determine the relationship between process variation, thereby changing the process.
Experimental design definition
- experimental design
Experimental design broad
- The broad design of experiments refers to the knowledge of the general procedures of scientific research. It includes a series of contents from the questioning, the formation of hypotheses, the selection of variables, etc., to the analysis of the results, and the writing of the thesis. It shows researchers an overview of how to carry out scientific research and tries to solve the whole process of research.
Experimental design narrow
- The narrow experimental design refers specifically to a plan for implementing experimental treatment and statistical analysis related to the plan.
- The narrow experimental design focuses on solving the paragraph from how to establish a statistical hypothesis to drawing conclusions.
Basic types of experimental design
- There are several basic types of commonly used psychological experiment designs, and these types are often used comprehensively.
- Single group design and contrast design
- Two basic design types based on whether a control group (control group) is set.
- experimental design
- Contrast design. This is one of the most basic designs of psychological experiments. It divided the subjects into two groups, one group was the experimental group, and the experimental treatment (also called treatment) was applied; the other group was the control group, without the experimental treatment. In order to make the two groups of subjects as homogeneous as possible for easy comparison, the random allocation method is generally used for grouping, and the effect of the experimental treatment is tested by measuring the difference between the two groups. Its basic mode is as follows.
- Two basic design types
- Two basic design types divided by grouping and processing.
- Completely randomized design. Also known as the test room design or independent group design. It originates from the sampling theory, that is, according to the principle of probability statistics, subjects are randomly assigned to each group and accept the treatment that each group should perform. Because it is randomly assigned, theoretically all aspects are equal before each group is processed. If the same treatment is applied to two or more groups under the same conditions, the average of the effects of each group should not be statistically significant; if two or more groups are treated differently, the resulting The difference in the average effect can be determined to be due to the difference in treatment. The experimental results of this design generally use independent t-test or analysis of variance. Modes III and IV listed in the comparative design can be said to be completely randomized designs, or they can be counted as comparative designs.
- experimental design
- Random block design. Also called subject design. It first divides the subjects into different groups according to certain characteristics, so that the subjects in each group are closer to homogeneity, and the subjects in the groups are more different. Then subjects in each block were randomly assigned to receive different treatments, or all treatments were received in different orders. In this way, all processing is accepted for one block. This is different from a completely randomized design. In a completely randomized design, each group only accepts the treatment that it should. V is the basic pattern of random block design. It differs from completely randomized design in that the variable block is also included in the experimental design. In this way, the total variation can be divided into "treatment room", "between blocks" and "error". Compared with completely randomized design, it can estimate the variation caused by individual differences. The basis for dividing blocks is closely related to the response variables to be examined, that is, when subjects in the same block are higher in the first experimental treatment than in other blocks, the score in the second treatment is also high. Therefore, the statistical method of random block design generally uses the t-test or analysis of variance of related samples. In addition, if each block in the random block design performs all the processing, it is called a complete block design; if the number of processes performed by each block is less than the total number of processes, it is called an incomplete block design . Although the latter does not perform all processing for each block, the number of blocks for each process must be the same. Most psychologists do not deal with much in the experiments, and basically use the complete block design. If the number of treatments is large (this is often encountered in agricultural experiments), due to the large number of experiments carried out, which is limited to manpower, financial resources and time, an incomplete block design must be used.
Experimental design factor design
- The two designs are divided according to the number of independent variables in the experiment.
- Single-factor design. There is only one independent variable, and other factors that can affect the results are controlled as unrelated variables. This design is concise and easy to use, but because there are often more than one factor that affects psychological activity in real life, when the situation is more complicated, it is best to use multi-factor experimental design.
- Multi-factor design. Independent variables are designed for two or more experiments. Commonly used multi-factor designs include complete randomization, random blocks, and Latin squares. The completely randomized multi-factorial design randomly groups according to the independent variables and the level of change (treatment) of each independent variable. In the 2 × 2 factor design, there are two independent variables, A and B, and each factor has two levels. There are 4 possible treatments, namely A1B1, A1B2, A2B1, and A2B2. It is necessary to randomly divide the subjects into 4 groups, and each group receives a treatment, that is, mode VI. Through 2 × 2 factor variance analysis of independent samples, the single effect of factor A or B and the interaction of A and B can be analyzed. Random block multi-factor design requires selecting a group of subjects in the 2 × 2 factor design, so that each subject receives 4 kinds of treatments, and which person receives which treatment in the order is determined randomly, so that each The result of the four kinds of processing of a person is a block. The statistical method used in this design is analysis of variance of related samples.
- The Latin square experimental design can accomplish the experimental purpose with fewer experiments. For example, there are 3 levels of A, B, and C factors, and 33 = 27 experiments are required. If you use the Latin square design, you do not have to do so many times. In this design, the arrangement of the two factors, A and B, is as follows.
- According to this arrangement, a total of 9 experiments. At the same time, C should be considered, and each level must be combined with different levels of the other two factors once, and the total number of experiments is still 9 times.
- Since this design often uses Latin letters, this arrangement is called Latin square. The statistical method is Latin square variance analysis. In addition to reducing the total number of experiments, this design also has the biggest advantage, which is that it can balance the effect of the experiment order. However, certain conditions must be met when using it, that is, it is assumed that there is no interaction between the factors, and the number of factors must be the same as the number of experimental treatment levels.
- experimental design
Quasi-experimental design
- A quasi-experimental design can be used when researchers have previously realized that some irrelevant variables will affect the experimental results but are actually difficult to properly control. The main feature of the quasi-experiment is that no randomization procedure is used, that is, the selection, grouping, and processing allocation of subjects are not randomly arranged.
- The main types of quasi-experimental designs
- Intermittent time series design: refers to the repeated observation of a group of subjects within a period of time before and after the treatment is performed, and the effect of the treatment is determined by comparing the observation results of the entire time series. The analysis of the obtained results requires testing and comparison of a series of observations before and after processing, and usually the relevant sample t-test method is used.
- Equal time sample design: refers to measuring a group of subjects in two equal periods of time, one of which is treated and the other is not treated, and then the observations obtained during the two equal periods are tested and compared
- experimental design
- Balanced adversarial design: It refers to providing control in the order of experimental processing to offset the sequence error caused by the influence of processing order. In a single-group experiment with only post-tests, if there are two treatments, A and B, the balanced adversarial design is performed in the order of ABBA.
Non-experimental design
- It is usually a natural description used to determine the critical variables that exist naturally and their relationships. There are many non-experimental research methods, such as natural observation method, correlation method, interview method, questionnaire method, test method, case method and biographical method. In non-experimental research, researchers neither use randomization procedures nor actively manipulate independent variables and control other unrelated variables.
- experimental design
- The main types of non-experimental design
- Single group post-test design: one test group and one post-test are given. Single group pre-test and post-test design: pre-test, processing and post-test of a test group. Neither study design has a control group. Static group comparison design: one of the two original groups (non-randomly selected) that had been organized before treatment was given treatment and post-test, while the other original group was not given treatment and only post-tested, and then Compare the results of the two groups. Retrospective design after the event: After the occurrence of an event, we started to collect various information about the event, and analyzed the effect and reason of the event. Because non-experimental design is a component or important element of true experimental design, it is also called pre-experimental design.
The main steps of experimental design
- The experiment plan and method strategy prepared by the researcher according to the research purpose before the experiment. Its main content is to reasonably arrange the experimental procedures, and to propose how to statistically analyze the experimental data and the main steps of psychological experiment design can be summarized as: put forward hypotheses according to the research purpose; draw up methods and procedures for verifying hypotheses; Statistical methods for processing and analyzing experimental data.
Experiment Design Activities
- It includes the following:
- 1. Establish statistical hypotheses related to research hypotheses;
- 2. Determine the experimental treatment (independent variables) used in the experiment and the additional conditions (extra variables) that must be controlled;
- 3 Determine the number of experimental units (subjects) required in the experiment and the total sample of the subjects;
- 4 Determine methods for assigning experimental conditions to participants;
- 5. Determine the measurement (dependent variable) and statistical analysis to be used for each subject in the experiment.
- Advantages of experimental design
- · Arrange experiments scientifically and reasonably, thereby reducing the number of experiments, shortening the experiment cycle, and improving economic benefits.
- · Identify the main factors that affect output from the many influencing factors.
- · Analyze the magnitude of interactions between influencing factors.
- · Analyze the impact of experimental errors to improve experimental accuracy.
- · Find the optimal parameter combination, and through analysis and comparison of the experimental results, find out the direction to achieve the optimal solution for further experiments.
- · Predict the output of the best solution.
Experiment design function
- experimental design
- The main function of the experimental design is to control the variables. First of all, the independent variables are effectively manipulated or changed under the control conditions, so that the changes of the dependent variables (that is, the response variables) can be observed. For example, when studying the impact of two teaching methods on children's academic achievement, the experiment designer should arrange to make other conditions as similar as possible, such as choosing two groups of children with similar home and school environments, similar academic foundations, and the same age. Different teaching methods, and then examine the impact of both on learning outcomes.
Experimental design
- Good experimental design mainly manifests in the reasonable arrangement of experimental procedures and effective control of irrelevant variables. Some irrelevant variables in psychological experiments can be eliminated by certain experimental instruments and techniques like physical and chemical experiments, but most of them are difficult to exclude, so they must rely on experimental design to balance or offset their effects. This control method is called experimental control method, and there are several commonly used:
- Elimination or keeping constant method: mainly use laboratory conditions to exclude the interference of irrelevant variables, and try to keep it constant for variables such as age, weight, experimental environment, and subject level that cannot be excluded;
- Balance method: the subjects are divided into experimental group and control group according to the principle of randomness, so that the effects of irrelevant variables on the two groups are equal;
- Cancellation method: its purpose is to control the influence caused by the experiment sequence, and the cyclic method is mainly used (the AB and BA methods are used when only two experiments are processed);
- Inclusion method: treat some irrelevant variable as an independent variable to make the experiment change from a single factor to a multifactorial design, and then conduct a multivariate statistical analysis of the results to find out the individual role and interaction of each independent variable. There are also some irrelevant variables. Although it is known that it has an effect on the results, it is impossible to use experimental control methods to balance or offset them. It can only be analyzed by statistical methods after the experiment is finished and excluded from the conclusion. This control method is called statistical control method. Commonly used statistical control methods are mainly covariance analysis or covariate analysis. This method is often used when research work cannot be randomly sampled on an individual basis due to practical difficulties or administrative reasons, and the integrity of the group must be maintained (for example, on a class basis).
- Experimental design evaluation criteria
- There can be many criteria for evaluating an experimental design, but it mainly depends on whether it can fully perform the following functions: properly solve the problem that the researcher wants to solve, that is, the experimental design must match the research problem; ", That is, it can effectively control irrelevant variables, so that the change of the response variable is completely determined by independent variables; The experimental results should be scientific and universal, and can be inferred to other subjects or other situations, that is, a higher" External validity. "
- Apply mathematical optimization theory and technology to experimental design, scientifically arrange experiments, and handle experimental results. The most effective technical method to use scientific methods to arrange experiments, process the test results, and obtain more and better production and scientific research results in the shortest time with the least human and material consumption.
The origin of experimental design optimization
- In the 1930s, due to the needs of agricultural experiments, RAFisher made a series of pioneering work in experimental design and statistical analysis, and since then experimental design has become a branch of statistical science. In the 1940s, during the Second World War, the U.S. military applied a large number of experimental design methods. Subsequently, F.Yates, RCBose, O. Kempthome, WGCochran, DRCox, and GEPBox all made outstanding contributions to the experimental design, making this branch more and more perfect in theory and more and more widely used. In the 1950s, Japanese statistician Genichi Taguchi tabularized the most widely used orthogonal design in experimental design, and made a well-known contribution to the wider use of experimental design in terms of method explanation.