非/参数检验

在进行ANOVA或Kruskal-Wallis检验后，若发现组间存在显著差异（拒绝原假设），通常需要进一步的事后检验（post-hoc tests）来确定具体是哪些组之间存在差异。选择哪种事后检验取决于数据的性质、假设条件以及研究设计。

• Purpose: After a significant ANOVA or Kruskal-Wallis test, post-hoc tests are used to determine which specific groups differ from each other.
  目的：在进行了显著的 ANOVA 或 Kruskal-Wallis 检验后，使用事后检验来确定哪些特定组之间存在差异。
• Examples: Bonferroni, Tukey's HSD, Dunn's test.
  例子：Bonferroni 检验、Tukey's HSD 检验、Dunn 检验。
• SPSS Procedures: Specified within the ANOVA or Kruskal-Wallis dialog boxes.
  SPSS 程序：在 ANOVA 或 Kruskal-Wallis 对话框中指定。
• Understanding: Adjusting for multiple comparisons to control the family-wise error rate.
  理解：调整多重比较以控制家族错误率

2. 分类变量的检验：

• Purpose: Determine if there is a statistically significant association between two or more categorical variables.
  目的：确定两个或多个分类变量之间是否存在统计学上的显著关联。
• Methods to Learn:  学习方法：
  • Chi-Square Test (χ²):  卡方检验（χ²）：
    • SPSS Procedure: Crosstabs (Analyze -> Descriptive Statistics -> Crosstabs). Make sure to select "Chi-square" in the "Statistics" options.
      SPSS 程序：交叉表（分析 -> 描述统计 -> 交叉表）。确保在“统计”选项中选择“卡方”。
    • Understanding: Expected vs. observed frequencies, degrees of freedom, p-value interpretation. Be aware of assumptions (e.g., expected cell counts).
      理解：预期频率与观察频率、自由度、p 值解释。注意假设（例如，预期单元格计数）。
  • Fisher's Exact Test:  Fisher 精确检验：
    • SPSS Procedure: Crosstabs. SPSS automatically calculates Fisher's Exact Test when one or more cells in a 2x2 table have an expected count less than 5.
      SPSS 程序：交叉表。当 2x2 表中有一个或多个单元格的预期计数小于 5 时，SPSS 会自动计算 Fisher 精确检验。
    • Understanding: Use when sample sizes are small or expected cell counts are low, better than Chi-square in these cases.
      理解：当样本量较小时或预期单元格计数较低时使用，这些情况下比卡方检验更好。
  • Yates's Correction for Continuity:
    Yates 连续性校正：
    • SPSS Procedure: Crosstabs. SPSS may apply Yates' correction automatically to 2x2 tables.
      SPSS 操作：交叉表。SPSS 可能会自动对 2x2 表格应用 Yates 校正。
    • Understanding: A modification of the Chi-square test for 2x2 tables to correct for the fact that the Chi-square distribution is continuous while the data are discrete.
      理解：对 2x2 表格的卡方检验的修正，以纠正卡方分布是连续的而数据是离散的事实。
• Key Concepts: Null hypothesis, alternative hypothesis, p-value, significance level (alpha), degrees of freedom, Type I and Type II errors.
  关键概念：原假设、备择假设、p 值、显著性水平（alpha）、自由度、I 类和 II 类错误。

3. Tests for Continuous Variables (Group Comparisons):
3. 连续变量检验（组间比较）：

• Purpose: Determine if there is a statistically significant difference in means or distributions of a continuous variable between two or more groups.
  目的：确定两个或多个组之间连续变量的均值或分布是否存在统计学上的显著差异。
• Methods to Learn:  学习方法：
  • Student's t-test (Independent Samples):
    t 检验（独立样本）:
    • SPSS Procedure: Compare Means -> Independent-Samples T Test (Analyze -> Compare Means -> Independent-Samples T Test).
      SPSS 操作：比较均值 -> 独立样本 T 检验（分析 -> 比较均值 -> 独立样本 T 检验）。
    • Understanding: Compares the means of two independent groups. Assumes data are normally distributed and have equal variances (Levene's test checks this).
      理解：比较两个独立组的均值。假设数据呈正态分布且具有相等方差（Levene 检验检查这一点）。
  • ANOVA (Analysis of Variance):
    方差分析（ANOVA）:
    • SPSS Procedure: Compare Means -> One-Way ANOVA (Analyze -> Compare Means -> One-Way ANOVA).
      SPSS 操作：比较均值 -> 单因素方差分析（分析 -> 比较均值 -> 单因素方差分析）。
    • Understanding: Compares the means of three or more groups. Assumes data are normally distributed and have equal variances (Levene's test checks this). Requires post-hoc tests (e.g., Bonferroni) for pairwise comparisons if the overall ANOVA is significant.
      理解：比较三个或更多组的均值。假设数据呈正态分布且具有相等方差（使用 Levene's 检验检查这一点）。如果整体 ANOVA 显著，则需要对数据进行成对比较（例如，使用 Bonferroni 进行事后检验）。
  • Non-parametric Alternatives (when assumptions of t-test or ANOVA are violated):
    非参数替代方法（当 t 检验或 ANOVA 的假设不成立时）：
    • Wilcoxon Rank-Sum Test (Mann-Whitney U test):
      Wilcoxon 秩和检验（Mann-Whitney U 检验）:
      • SPSS Procedure: Nonparametric Tests -> Legacy Dialogs -> 2 Independent Samples (Analyze -> Nonparametric Tests -> Legacy Dialogs -> 2 Independent Samples). Select Mann-Whitney U.
        SPSS 操作：非参数检验 -> 传统对话框 -> 两个独立样本（分析 -> 非参数检验 -> 传统对话框 -> 两个独立样本）。选择 Mann-Whitney U。
      • Understanding: Compares the distributions of two independent groups. Does not assume normality.
        理解：比较两个独立组的分布。不假设正态性。
    • Kruskal-Wallis Test:  Kruskal-Wallis 检验:
      • SPSS Procedure: Nonparametric Tests -> Legacy Dialogs -> K Independent Samples (Analyze -> Nonparametric Tests -> Legacy Dialogs -> K Independent Samples).
        SPSS 过程：非参数检验 -> 传统对话框 -> K 个独立样本（分析 -> 非参数检验 -> 传统对话框 -> K 个独立样本）。
      • Understanding: Compares the distributions of three or more groups. Does not assume normality. Requires post-hoc tests for pairwise comparisons if the overall test is significant.
        理解：比较三个或更多组的分布。不假设正态性。如果整体检验显著，需要事后检验进行配对比较。

⛄⛄⛄⛄⛄⛄⛄⛄⛄⛄⛄⛄⛄⛄⛄⛄⛄⛄⛄⛄⛄⛄⛄⛄⛄⛄⛄⛄⛄⛄⛄

Okay, let's break down the premise and data usage for each statistical test in the paper, keeping in mind that these are general guidelines and the authors would have (or should have!) checked the assumptions for their specific data.
好的，让我们分析论文中每个统计测试的前提条件和数据使用情况，记住这些是一般性指南，作者们应该（或者应该！）检查了他们特定数据的假设。

1. Fisher's Exact Test:  1. Fisher 精确检验：

• Premise (When to Use):  前提（何时使用）：

  • Two Categorical Variables: You want to see if there's an association between two variables where each variable has distinct categories (e.g., gender: male/female; recurrence: yes/no).
    两个分类变量：你想查看两个变量之间是否存在关联，其中每个变量都有不同的类别（例如，性别：男/女；复发：是/否）。
  • Small Sample Size or Low Expected Cell Counts: This is the key condition. You use Fisher's when the total sample size is small or when, in your contingency table, at least one cell has an "expected count" less than 5. An "expected count" is what you'd expect to see in a cell if there was no relationship between the variables. SPSS will usually flag this for you.
    样本量小或期望单元格计数低：这是关键条件。当总样本量较小时，或在你列联表中，至少有一个单元格的“期望计数”小于 5 时，你使用 Fisher 精确检验。“期望计数”是在变量之间没有关系时，你期望在单元格中看到的结果。SPSS 通常会为你标记这一点。
  • Independent Observations: The observations (patients) should be independent of each other.
    独立观察：观察结果（患者）应相互独立。

• Data Usage (In This Paper - Examples):
  数据使用（本文示例）：

  • Gender and Recurrence/Malignant Transformation: Did being male or female influence the likelihood of recurrence?
    性别与复发/恶性转化：男性或女性身份是否影响了
  • LPR (Laryngopharyngeal Reflux) and Recurrence/Malignant Transformation: Did having LPR influence the likelihood of recurrence?
    LPR（喉咽反流）和复发/恶性转化：LPR 是否影响复发的可能性？
  • Size of Lesion and Recurrence/Malignant Transformation: Was the size of the lesion (<50% vs. >=50%) related to recurrence?
    病灶大小和复发/恶性转化：病灶大小是否小于 50%？
  • Site of Lesion and Recurrence/Malignant Transformation: Was a unilateral or bilateral lesion more likely to recur?
    病灶部位和复发/恶性转化：单侧或双侧病灶更可能复发？

2. Yates's Correction for Continuity:
2. 耶茨连续性校正：

• Premise (When to Use):  前提（何时使用）：

  • Two Categorical Variables (2x2 Table): Similar to Chi-square, but specifically for 2x2 contingency tables (two categories for each variable).
    两个分类变量（2x2 表格）：类似于卡方检验，但专门用于 2x2 列联表（每个变量有两个类别）。
  • Small Sample Size (Sometimes): Yates' correction is an adjustment to the Chi-square test to make it more accurate when dealing with small sample sizes in 2x2 tables. It helps correct for the fact that the Chi-square distribution is continuous, while the data are discrete.
    小样本量（有时）：Yates 校正是对卡方检验的调整，以提高其准确性

• Data Usage (In This Paper):
  数据使用（在本论文中）：

  • It would be used in exactly the same situations as Fisher's exact test if the researchers chose to use a corrected Chi-square instead of Fisher's. However, Fisher's exact is generally preferred over Yates' correction when expected cell counts are low.
    如果研究人员选择使用校正的卡方检验而不是费希尔精确检验，它将用于与费希尔精确检验相同的情况。然而，当期望细胞计数较低时，通常更倾向于使用费希尔精确检验而不是耶茨校正。

3. Pearson's Chi-Square Test:
3. 皮尔逊卡方检验：

• Premise (When to Use):  前提（何时使用）：

  • Two Categorical Variables: Same as Fisher's, you're looking for an association.
    两个分类变量：与费希尔相同，你正在寻找关联。
  • Larger Sample Size and Adequate Expected Cell Counts: This is the key difference. The Chi-square test is reliable when you have a larger sample size and the expected cell counts in your contingency table are generally 5 or greater.
    较大的样本量和充足的预期单元格计数：这是关键区别。当您有较大的样本量，并且您的列联表中的预期单元格计数通常为 5 或更大时，卡方检验是可靠的。
  • Independent Observations.
    独立观察。

• Data Usage (In This Paper - Examples):
  数据使用（本文示例）：

  • Assuming the sample size requirements were met, would be identical to Fisher's and Yates' scenarios.
    如果样本量要求得到满足，将与 Fisher 和 Yates 的场景相同。

4. Wilcoxon Rank-Sum Test (Mann-Whitney U Test):
4. Wilcoxon 秩和检验（Mann-Whitney U 检验）:

• Premise (When to Use):  前提（何时使用）：

  • One Continuous/Ordinal Variable, One Categorical Variable (Two Groups): You want to compare the distributions of a continuous or ordinal variable between two independent groups.
    一个连续/有序变量，一个分类变量（两组）：你想比较两个独立组之间一个连续或有序变量的分布。
  • Non-Normal Data (Often): This is the key reason to use Wilcoxon Rank-Sum. You use it when the data don't meet the assumption of normality required for a t-test. The data may be skewed or have outliers. Wilcoxon Rank-Sum is a non-parametric test, meaning it doesn't rely on assumptions about the shape of the distribution.
    非正常数据（通常）：这是使用 Wilcoxon Rank-Sum 的关键原因。当数据不符合 t 检验所需的正态性假设时，您使用它。数据可能存在偏斜或有异常值。Wilcoxon Rank-Sum 是一种非参数检验，意味着它不依赖于关于分布形状的假设。
  • Ordinal Data: The variable might be ordinal (e.g., a rating scale from 1 to 5).
    序数数据：该变量可能是序数（例如，从 1 到 5 的评分量表）。

• Data Usage (In This Paper - Examples - Hypothetical):
  数据使用（在本论文中 - 示例 - 假设的）

  • The paper doesn't explicitly state using this test, but it's possible they used it if data was not normally distributed
    这篇论文没有明确说明使用这个测试，但如果数据不是正态分布的，他们有可能使用了它
  • Smoking Status (Ordinal Categories) and Recurrence: Does the level of smoking (non-smoker, former smoker, current smoker) influence the likelihood of recurrence. Because these are ordered categories, it is a continuous variable, and the Wilcoxon can apply.
    吸烟状况（有序类别）和复发：吸烟程度（非吸烟者、前吸烟者、现吸烟者）是否影响复发的可能性。因为这些是有序类别，所以它是一个连续变量，可以使用 Wilcoxon 检验。

5. Student's t-Test (Independent Samples):
5. 学生 t 检验（独立样本）：

• Premise (When to Use):  前提（何时使用）：

  • One Continuous Variable, One Categorical Variable (Two Groups): You want to compare the means of a continuous variable between two independent groups.
    一个连续变量，一个分类变量（两组）：你想比较两个独立组之间连续变量的均值。
  • Normality: The data in each group should be approximately normally distributed. This can be checked with histograms, Q-Q plots, or normality tests (Shapiro-Wilk, Kolmogorov-Smirnov).
    正态性：每个组中的数据应近似正态分布。这可以通过直方图、Q-Q 图或正态性检验（Shapiro-Wilk 检验、Kolmogorov-Smirnov 检验）来检查。
  • Equal Variances (Often): Ideally, the two groups should have approximately equal variances. Levene's test can be used to check for equality of variances. If variances are not equal, you can use a modified t-test (Welch's t-test) that doesn't assume equal variances.
    方差齐性（通常）：理想情况下，两组应具有近似相等的方差。可以使用 Levene 检验来检查方差的齐性。如果方差不相等，可以使用不假设方差相等的修正 t 检验（Welch t 检验）。
  • Independent Observations.
    独立观察。

• Data Usage (In This Paper - Examples):
  数据使用（本文示例）：
  The H-SCORE different between the recurrence and nonrecurrence groups was statistically significant (126.3 ± 56.8 vs. 78.4 ± 49.9, P = 0.005, Fig. 3 and Supplemental Table 2). Similarly, the H-SCORE different between the malignant transformation group and the non-malignant transformation group was also statistically significant (137.5 ± 74.3 vs. 87.8 ± 47.3, P = 0.045, Fig. 3 and Supplemental Table 3)."
  复发组和非复发组的 H-SCORE 差异具有统计学意义（126.3 ± 56.8 vs. 78.4 ± 49.9，P = 0.005，图 3 和补充表 2）。同样，恶变组和非恶变组的 H-SCORE 差异也具有统计学意义（137.5 ± 74.3 vs. 87.8 ± 47.3，P = 0.045，图 3 和补充表 3）。

  Here, the independent t-test may have been used to compare:
  这里，独立样本 t 检验可能已被用于比较：
  * The difference in H-SCORE levels between the recurrence and nonrecurrence VFL groups
  * 复发组和非复发组 VFL 组之间 H-SCORE 水平的差异

Key Takeaways:  关键要点：

• Assumptions Matter: The validity of these tests depends heavily on meeting their underlying assumptions. It's the researcher's responsibility to check these assumptions and choose the appropriate test.
  假设很重要：这些测试的有效性很大程度上取决于满足其基本假设。检查这些假设并选择适当的测试是研究者的责任。
• Data Type Drives the Choice: The type of data you have (categorical vs. continuous/ordinal) is the biggest factor in deciding which test to use.
  数据类型决定了选择：你所拥有的数据类型（分类与连续/有序）是决定使用哪个测试的最大因素。
• SPSS Can Help: SPSS can calculate these tests and provide you with the statistics needed to interpret the results (p-values, test statistics, etc.). It can also help you check some of the assumptions (e.g., Levene's test for equality of variances).
  SPSS 能帮助您：SPSS 可以计算这些检验并提供您解释结果所需的统计数据（p 值、检验统计量等）。它还可以帮助您检查一些假设（例如，Levene 方差齐性检验）。

It's crucial to consult a statistician or refer to a good statistics textbook for more detailed information on these tests and their assumptions. This overview is intended to provide a general understanding.
重要的是咨询统计学家或参考一本好的统计学教科书，以获取这些测试及其假设的更详细信息。本概述旨在提供一般理解。