Reliability analysis using SPSS software

Reliability Analysis of questionnaires First, the concept:Reliability is the index that reflects the true degree of the measured feature according to the consistency or stability of the results obtained by the test tool. Generally speaking, the results of two or two Tests are the same, the smaller the error, the higher the reliability, which has the following characteristics:

1. Reliability refers to the consistency or stability of the results obtained by the test, rather than the test or scale itself;

2, the reliability value refers to in a certain type of consistency, not the general consistency, the reliability coefficient will be different time, different subjects or different raters to appear different results;

3, reliability is the necessary condition of validity, not sufficient condition. Reliability is low, but reliability is not high;

Reliability testing relies entirely on statistical methods.

Reliability can be divided into: intrinsic reliability: Whether a group of questions is measured by the same concept, and the degree of internal consistency of the composition of the items; The common detection method is the Cronbach ' s alpha coefficient. External reliability: The test results for the same testers at different times are consistent, and retest reliability is the most common test method for external reliability.

second, the Reliability index:

1. The reliability coefficient is used to indicate the degree of reliability. The greater the reliability coefficient, the greater the credibility of the measurement. There is a high reliability of the reliability coefficient. Scholar Devellis (1991) believed that 0.60~0.65 (preferably not); 0.65~0.70 (minimum acceptable value); 0.70~0.80 (very good); 0.80~0.90 (very good). Therefore, a good reliability of the scale or questionnaire, preferably in more than 0.80, between 0.70 to 0.80 is acceptable range; the component table is preferably above 0.70, between 0.60 to 0.70 acceptable. If the internal consistency coefficient of the component table is below 0.60 or the reliability coefficient of the total table is below 0.80, it should be considered to revise the scale or additions and deletions.

2. The reliability index is expressed by correlation coefficient: it can be divided into three types: stability coefficient (consistency of span time), equivalence coefficient (cross form consistency) and intrinsic consistency coefficient (cross project consistency). Third, the Reliability analysis method:

1. Retest Reliability Method:

The same questionnaire is used to repeat the test of the same interval for a certain period of time, also known as the test-re-test method, to calculate the correlation coefficient of the two results. Obviously this is the stability factor, that is, the consistency across time. The retest reliability method is applicable to the factual questionnaire, and also can be used for attitude and opinion questionnaire which is not easy to be affected by the environment. Because the retest reliability needs to test the same sample two times, and the test is susceptible to various events, activities, so interval time need to be appropriate. More often than not, the interval is two weeks or one months.

2. Copy Reliability method (equivalent reliability method):

The copy reliability method is to allow two copies of questionnaires to be completed at one time and to calculate the correlation coefficients of two copies. Since this method requires the two copies in addition to different expressions, in terms of content, format, difficulty and the question of the corresponding questions, such as the way to be completely consistent, so the copy reliability is equivalent coefficient. In the actual survey, the questionnaire is difficult to meet this requirement, this method is less used.

3. Binary Reliability Method:

Binary reliability method means that the measurement items are divided into two halves according to the odd and even items, respectively, and the correlation coefficients between the two halves (the actual application of Excel software) are calculated, and then the reliability coefficient of the whole measurement rxx is determined. Binary reliability belongs to intrinsic consistency coefficient, and it measures the consistency between two halves of the project. This method is not suitable for the measurement of factual questionnaires, and is often used in the reliability analysis of attitude and opinion questionnaires. In the questionnaire survey, the most common form of attitude measurement is the level 5 Likert scale. For binary reliability analysis, if the scale contains an anti-Italian title, should be the inverse of the goal of the item to deal with the score in order to ensure consistency in the direction of the points, and then all the items according to the odd 偶或 before and after divided into the same as possible two halves, to calculate the correlation coefficient (RHH), that is half the scale of the reliability coefficient), Finally, the Spearman-Brown (Spearman-brown) formula: rtt=2rhh/(1+RHH) is used to find the reliability coefficient rtt of the whole scale.

4. Rating Reliability:

This method is carried out in a low degree of standardization of measuring tools. The assignment standard of the different raters also affects the reliability of the measurement, and to test the reliability of the raters, one can calculate the correlation coefficient of a group of scores with another score.

5. A reliability coefficient method:

Krumbach a reliability coefficient is the most commonly used reliability coefficient at present. The formula is: a= (k/k-1) * (∑SI2)/st2)
Among them, K is the total number of the items in the scale, SI2 is the variance of the problem in the first question, ST2 is the variance of the total score of all the titles. It can be seen from the formula that the evaluation of a coefficient is the consistency between the scores of each item in the scale, which belongs to the intrinsic consistency coefficient. This method is applicable to the reliability analysis of attitude and opinion Questionnaire (scale).

The reliability test method used in Likert scale method is "Cronbach ' s a" coefficient and "binary reliability".

iv. Reliability Analysis using SPSS

In SPSS, the module which is used to analyze the test reliability is reliability analysis under scale, and the factor module under the data reduction is used.

The main function of reliability Analysis module is the reliability of test test, which is mainly used to test the binary reliability, Ku Li and a coefficient and Hoyt reliability coefficient. As for the retest reliability and the reliability of the replica, only after merging the data from the scores of the two test points into the same data file, the correlation coefficients of the correlate under the bivariate are obtained, that is, the retest or the copy reliability And the reliability of the raters is to use the Spearman grade correlation and the Kendall harmony coefficient.

Table 1 parameters of model options for Reliability Analysis module and corresponding Chinese terminology

Key words	Function
Alpha	Cronbach a coefficient
Split-half	Binary reliability, N is the number of questions in the second Component table
Guttman	Guttman minimum lower bound true reliability method
Parallel	The maximum probability (maximum-likelihood) reliability of the homogeneity of variance numbers in different subjects
Strict Parallel	The maximum probability reliability of the average and variance numbers of each topic being homogeneous

Table 2Reliability Analysis module of the statistics part of the parameters of the options and corresponding Chinese terminology

Keyword	Features
F test	Hoyt Reliability factor
Friedman Chi	Friedman rating Variance analysis and Kendall Harmony factor
Cochran Chi	Cochran ' s Q Test, applicable to scale with answer two (for example, right)
Hotelling ' s T	Hotelling ' s T2 check
Tukey ' s	Tukey extensibility Check
Intraclass	The average number of questions in the list of issues