First, the concept (analysis-regression-ranking)

Ordered regression can be used to simulate the dependence of multivariate ordered response on a set of predictive variables (which can be factors or covariates). The design of ordered regression is based on the methodology of McCullough (1980, 1998). The process of ordinal regression is grammatically called plum blossom. For example, ordered regression can be used to study patients' response to drug doses. Possible reactions can be classified as none, mild, moderate or severe. The difference between mild reaction and moderate reaction is difficult or impossible to quantify, which depends on the feeling. In addition, the difference between mild reaction and moderate reaction may be greater or smaller than the difference between moderate reaction and severe reaction.

II. Options (Analysis-Regression-Sorting-Options)

Using the Options dialog box, you can adjust the parameters used in the iterative estimation algorithm, select the confidence of the estimated values of the parameters, and select related functions.

1, iteration. You can customize the iterative algorithm. ◎ Maximum iteration times. Please specify a non-negative integer. If you specify 0, the procedure returns an initial estimate. ◎ The maximum step length is divided into two parts. Please specify a positive integer. ◎ Convergence of log-likelihood estimation. If the absolute or relative change of log-likelihood estimation is less than this value, the algorithm will stop. If you specify 0, the condition is not used. ◎ Parameter convergence. If the absolute or relative change of the estimated value of each parameter is less than this value, the algorithm will stop. If you specify 0, the condition is not used.

2. Confidence interval. Please specify a value greater than or equal to 0 and less than 100.

3. Delta. The value added to the zero cell frequency. Please specify a non-negative value less than 1.

4. Singularity tolerance error. Used to check highly correlated predictive variables. Select a value from the list of options.

5. Link function. Link function is a transformation form of cumulative probability, which can be used for model estimation. The following table summarizes the five available link functions. ◎Logit log (? / ( 1? )) evenly distributed categories. ◎ Complementary log (log( 1? )) The higher the category, the greater the possibility. ◎ negative logarithm log(log (? )) The lower the category, the greater the possibility. ◎Probit？ 1(? ) The latent variables are normally distributed. ◎ Cauchy (inverse Cauchy) tan(π (? 0.5)) Potential variables have many extreme values.

Thirdly, orderly regression output (analysis-regression-sequence-output)

The Output dialog box can generate a table to be displayed in the browser and save the variables in the working file.

1, display. Generate tables for the following items: ◎ Print iteration history. Print the log-likelihood estimation and parameter estimation of the specified printing iteration frequency. Always print the first and last iteration. ◎ goodness of fit statistics. Pearson and Likelihood Chi-Square Statistics. These statistics are calculated according to the categories specified in the variable list. Abstract statistics. Cox and snell, nagel Kirk and mcfadden R2 statistics. ◎ Parameter estimation. Parameter estimation, standard error and confidence interval. ◎ Asymptotic correlation of parameter estimation. Correlation coefficient matrix of parameter estimation. ◎ Asymptotic covariance of parameter estimation. Covariance matrix of parameter estimation. Cell phone information. Observed and expected frequency and cumulative percentage, Pearson residual of frequency and cumulative percentage, observed and expected probability, and cumulative probability of each response category expressed in covariant mode. Note: For models with many covariant patterns (for example, models with continuous covariant), this option may generate very large and difficult tables. ◎ Parallel test. Hypothesis test of equality of horizontal position parameters of multiple dependent variables. This test is only applicable to positioning models only.

2. Saved variables. Save the following variables in the working file: ◎ Estimate the response probability. Model estimation probability of classifying factor/covariant patterns into response categories. The probability is equal to the number of response categories. ◎ Prediction class. Response category with maximum estimated probability of factor/covariant mode. ◎ Predicted category probability. The estimated probability of classifying factors/covariates into prediction categories. This probability is also the maximum of the estimated probability of the factor/covariant mode. ◎ Actual category probability. The estimated probability of classifying factors/covariates into actual categories.

3. Possibility of printing logs. Controls the display of log-likelihood estimation. ◎ Including polynomial constants can provide the complete value of likelihood estimation. To compare results between products that do not contain this constant, you can choose to exclude this constant.