Received March 22, 2012; accepted March 25, 2012; published online May 15, 2011.
Full-text LinkOut at PubMed. Journal title in PubMed: Zhong Xi Yi Jie He Xue Bao.

基金项目：国家科技重大专项“重大新药创制”资助项目 (No. 2009ZX09502-028)
Correspondence: Prof. Liang-ping Hu; Tel: 010-66931130; E-mail: lphu812@sina.com

     The balance principle is the kernel of the four principles, balance, randomization, control and repetition, which plays a guiding role in scientific research. It means that among different groups of the same experimental factor, all the non-experimental factors are balanced^[1].

1  Function and value of the balance principle
     The function of the balance principle is to balance the influence of non-experimental factors on research subjects among different groups so that the influence of different levels of the examined experimental factor will be truly revealed^[1-3]. For instance, a researcher wanted to conduct an experiment to examine whether bread containing lysine would help the growth of adolescence. He chose a class of students in a primary boarding school as the research subjects, and randomly divided the students into two groups. The students in one group had a loaf of bread every morning and afternoon respectively during the break, while the students in the other group did not. One year later, the researcher compared the physical development of the two groups of students and drew the conclusion that lysine could help the adolescence grow better. The conclusion is unreliable and even ridiculous because during the break, the students may feel hungry so that even having a loaf of bread not containing lysine may help their growth. That is to say, the two groups are not balanced on the important non-experimental factors (whether or not having a loaf of bread every morning and afternoon), which causes the conclusion unreliable. The appropriate way is to let the students in one group have a loaf of bread containing lysine every morning and afternoon during the break, while let the students in the other group have a loaf of bread not containing lysine every morning and afternoon during the break. The two kinds of bread should be same in size and ingredient. Besides, the two groups of students should be balanced on other important non-experimental factors, for instance, exercise, sleeping hours, etc. One year later, when the researcher compares the physical development of the two groups, he may able to know the value of lysine in the growth of adolescence.
     The balance principle seems to be unrelated to the other three principles in scientific research. In fact, it is not only closely related to them, but also plays a leading role in the success of scientific research. When drafting a research plan, if the balance principle is well considered, the research plan will be scientific and perfect; on the other hand, if the balance principle is not well considered, the research plan may lead the whole research to a complete failure. Therefore, if we say the three elements, the four principles and the design type are the essence of a research design, then the balance principle is the kernel of the essence.

2  Strategies to improve balance in scientific research

     The strategies to improve balance in scientific research can be generalized into the following three aspects. First, the researcher should know his specialty and statistics well, and has the ability to lay down an ideal research plan. Second, after drafting the research plan, the researcher should invite experienced experts on his specialty to review the research plan and ask for their opinions and advice. Third, the researcher should invite experienced experts on statistics to review the research plan and ask for their advice. It is better if the experts on statistics participate in drawing up the research plan^[4].

3  Detailed measures to improve balance in scientific research

     Before conducting an experiment, a researcher should draw up a scientific research plan. That is to say, the three elements, the four principles, the research type and the quality control should be well considered.
     How to have a good command of the three elements? In terms of research subjects, the researcher should lay down rational inclusion and exclusion criteria for research subjects. In terms of research factors, the researcher should find the experimental factors and their levels correctly and should pay much attention to finding the important non-experimental factors which may influence a lot on the result to make sure that they are balanced among groups. In terms of observed indexes, the researcher should differentiate between the major and minor indexes of efficacy and safety.
     How to have a good command of the four principles? It is important that the researcher adheres to the principles of randomization, control, repetition and balance and strictly performs them. The principle of randomization requires that the research subjects are assigned based on the important non-experimental factors following the stratified randomization. In clinical research, sometimes the research subjects can be divided based on the dividing principle of the minimum unbalanced index following the dynamic randomization. The principle of control emphasizes on the rational setting of control groups. The principle of repetition requires that the sample size should be large enough under specific design type. It is strongly recommended that the sample size is estimated according to the right formula. As long as the above three principles are strictly followed, the balance principle will then be guaranteed.
     How to have a good command of the design type? The researcher should choose the right design type based on the number of the experimental factors and their levels, the block factors (the important non-experimental factors) and the specialty. The multifactorial design of a certain kind of type is recommended if possible.
     How to have a good command of quality control? The reliability of a research result is closely dependent on many aspects, for instance, the researchers, the research subjects, the environment, the research conditions, etc. Each aspect should be paid enough attention to. The researchers should take strict skill training and pass the exams so that they are qualified both psychologically and technically. The research subjects should be managed to lower the interference of psychological factors with the result. The researcher should try to improve the compliance of the research subjects. In addition, the experimental conditions should be under standardized control, for example, the instruments and reagents should be correctly calibrated, and the criteria for diagnosis, measurement and evaluation should be scientific and unified.
     During statistical analysis, the researcher should use as much information as possible and adopt appropriate statistical methods to eliminate the influence of the confounding factors on the result.

4  Common mistakes involving the balance principle in scientific research and analyses

4.1  Example 1  In order to examine the efficacy of a certain compound traditional Chinese medicine on angina pectoris of coronary heart disease, 136 patients who were diagnosed angina pectoris of coronary heart disease were chosen as the research subjects and were divided into two groups: the experimental group and the control group. The experimental group consisted of 101 patients, of which 55 were male and 46 were female, with an average age of (64.0±5.7) years old. Among the 101 patients, 90 suffered from stable angina pectoris and 11 had unstable angina pectoris; 62 patients complicated with hypertension; 42 complicated with hyperlipidemia, 13 complicated with diabetes and 15 complicated with old myocardial infarction. The control group consisted of 35 patients, of which 17 were male and 18 were female, with an average age of (63.0±5.3) years old. Among the 35 patients, 33 suffered from stable angina pectoris and 2 suffered from unstable angina pectoris; 21 complicated with hypertension, 13 complicated with hyperlipidemia, 2 complicated with diabetes and none complicated with old myocardial infarction. Question: Are the experimental group and the control group balanced in this clinical trial?
     The balance principle requires that different groups of the same experimental factor should be balanced on all the non-experimental factors. According to our common sense, the type of angina pectoris and the complications will interfere with prognosis. The researcher should combine his expertise and statistical knowledge to balance the experimental group and the control group in terms of the disease types and complications during the experimental design phase so that the conclusion will be persuasive. In this example, the following measures may be adopted to improve the balance between the two groups. First, the type of angina pectoris (stable or unstable) and the complications (hypertension, hyperlipidemia, diabetes or old myocardial infarction) form eight groups in a complete combination; then randomly divide the patients in each group into the experimental group and the control group so that the type of angina pectoris and the complications will be balanced between the experimental group and the control group. It is called the stratified randomization based on different types of angina pectoris and different complications, which demonstrates that the type of randomization has a direct influence on the balance of an experimental design.
4.2  Example 2  To observe the effects of Chinese herbal medicine of nourishing yin and clearing heat on vascular endothelial growth factor (VEGF) in growth retardation induced by decamethasone and observe its mechanisms, 31-month-old New Zealand white rabbits were randomly divided into control group, dexamethasone-treated group and Chinese herbal medicine-treated group. The rabbits in the dexamethasone-treated group and the Chinese herbal medicine-treated group received dexamethasone (5 mg/kg per day). The rabbits were sacrificed at the 6th and 12th weeks after administration, and then the tibia articular was removed. The TUNEL staining was used to detect apoptotic index. The immunohistochemical staining was used to detect the positive index of the expression of VEGF in the epiphyseal cartilage of growth. The fluorescent quantitative polymerase chain reaction was adopted to detect the expression intensity of VEGF mRNA in each group. The result was listed in Table 1 as follows ^[5]. Question: Is the design correct?

Table 1  The influence of Chinese herbal medicine on VEGF in the epiphyseal cartilage of rabbit tibia

（x±s）

VEGF: vascular endothelial growth factor.

     Example 2 involves three experimental factors: “using dexamethasone or not”, “using the Chinese herbal medicine or not” and “time”, each of which has two levels. Therefore, there are supposed to be eight different experimental conditions if the three experimental factors are in a complete combination. However, Table 1 only presents six experimental conditions. The other two groups “using the Chinese herbal medicine only at the 6th week and the 12th week” are missing, which causes the mistake of incomplete control, and thus it violates the balance principle.
     The appropriate way to conduct the experiment is to adopt the three-factor factorial design so that the experimental effect of each factor will be revealed, and the interactions among the three factors will be shown. Table 2 presents the right form of Table 1.

Table 2  The influence of Chinese herbal medicine on VEGF in the epiphyseal cartilage of rabbit tibia (right form)

（x±s）

VEGF: vascular endothelial growth factor.

     The three experimental factors are clearly shown in Table 2. If the design was consummate, the 2nd line in Table 2 would have the corresponding data. If the three indexes are independent, the univariate analysis of variance of quantitative data with the three-factor factorial design can be adopted to examine the effects of the three experimental factors and their interactions on each index; if the three indexes are related according to the specialty and should be examined together, the three-variant analysis of variance of quantitative data with the three-factor factorial design can be adopted to examine the effects of the three experimental factors and their interactions on the three indexes.