How should the data of SCI papers be analyzed? What are the methods? In this issue, the editor has sorted out some data analysis methods of SCI papers, hoping to help you.

　　The sensitivity analysis of SCI papers is to evaluate the robustness of methods by changing methods, models, unmeasured variable values and assumptions (when the assumptions and methods of data analysis are changed, the results are consistent with the main conclusions). In essence, it is to change the assumed conditions and statistical methods to conduct statistical analysis again to determine whether the results have changed. The purpose is to investigate the stability of the results and improve the credibility of the conclusions.

　　1、 Applicable to which scenarios

　　1. Data

　　2. Analysis population

　　3. Variable definition

　　4. Statistical model

　　5. Distribution assumption

　　2、 What are the presentation forms

　　1. Statistical chart (rich in forms and colors)

　　2. Statistical tables (easy to make and provide accurate data)

　　3. Figure + table

　　(1) Statistical chart:

　　The results of sensitivity analysis and main analysis results are integrated into a statistical map of effect estimation, which is common in forest map with subgroup analysis.

　　The results of sensitivity analysis and main analysis are integrated into a statistical table of effect estimation, which is common in varying degrees of covariate correction.

　　(2) How to reflect the results of sensitivity analysis?

　　Sensitivity analysis result = main analysis result [brief description] as attached table or figure

　　Sensitivity analysis results ≠ main analysis results [analysis and interpretation] directly present the sensitivity analysis results

　　3、 What are the commonly used statistical methods in the data?

　　(1) Outliers 1. Hypothesis test 2. Labeling method

　　① Standard deviation method (>3*sd) ② Z-value method (absolute value >3)

　　③ Improved Z-value method (median, mad)

　　④ Box diagram (> more than twice the interquartile spacing (q3-q1))

　　⑤ Compare the outliers marked by the inclusion and elimination tagging methods

　　⑥ Robust regression (least median square (LMS) method, M-estimation method)

　　(2) Missing value

　　1. Complete data method (applicable: large sample size, less proportion of missing values (<5%))

　　2. Missing data tagging

　　3. Missing value interpolation

　　① Single value interpolation ((advantage: simple and easy to operate)

　　Single value interpolation methods include mean value (for normal continuous variables), median (partial continuous variables), mode (classified variables) interpolation last data truncation method, best data truncation method, and worst data truncation method. Recommended reading: fake papers

　　② Multiple interpolation

　　Including linear regression, prediction mean matching, propensity score, logistic regression, discriminant function, Markov chain Monte Carlo, full condition definition.

　　4、 What are the commonly used statistical methods in the statistical model?

　　(1) Group effect

　　Objective: to control the group effect.

　　Commonly used generalized estimation equation (GEE), mixed effect model II, competitive risk model

　　① The cumulative incidence rate (CIF) of each event was estimated, and gray's performed the inter group difference test;

　　② Cause specific risk function and partial distribution risk function.