MR

Week_11_Lecture_Recording

Overview of Chi-Square and Log-Linear Analysis

  • Focus on tests of association using frequencies

  • Key Applications:

    • Analyze relationships between categorical variables

    • Utilize SPSS for computations and output

    • Explore effect sizes and odds ratios as interpretations of statistical significance

Categorical Data Analysis

  • Categorical data can be assessed using:

    • Chi-square tests

    • Log-linear models

  • Examples of categorical responses can include:

    • Voting frequency for politicians

    • Students passing or failing subjects

    • Patients diagnosed or free from diagnosis after treatment

  • Caution in analysis: Numeric values of categories are arbitrary; means are meaningless for categorical variables.

Chi-Square Test Basics

  • Pearson Chi-Square:

    • Used to assess the relationship between two categorical variables

    • Compares observed frequencies to expected frequencies to determine if deviations are due to chance

  • Example Scenario:

    • Analyzed training methods (food vs affection) on whether cats can learn to dance.

    • Utilizing a contingency table to visualize: training methods vs dance outcome (yes/no).

  • Equation for Chi-Square:

    • [ \chi^2 = \sum \frac{(O_{ij} - E_{ij})^2}{E_{ij}} ] where:

      • (O_{ij}) = observed data frequencies

      • (E_{ij}) = expected frequencies based on chance

      • Rows and columns represent categories in the contingency table

  • Key Terms:

    • Observed Frequencies: Real counts from the experiment

    • Expected Frequencies: Counts expected if there were no association

  • Use of standardized residuals to interpret results:

    • Standardized residuals are akin to z-scores, indicating significance at p < 0.05 if outside the range of +/- 1.96.

Log-Linear Models

  • Applied when assessing association among three or more categorical variables.

  • An example using dogs and cats, allowing for different predictor variables while analyzing training effects.

  • Key Concepts:

    • Models assess main effects and interaction effects (e.g., animal type, training method, and dance outcome).

    • Uses backward elimination to simplify models, assessing significance of interactions and removing non-significant terms.

  • Importance of assessing the k-way interactions and their individual main effects:

    • Statistical significance checks if removing interactions shortens the model significance.

Assumptions for Chi-Square Tests

  1. Independence: Each data point contributes uniquely to contingency cells.

  2. Expected Counts: No cell should have an expected frequency less than 5; overall, fewer than 20% of cells should meet this condition in larger tables.

  3. Small samples may require Fishers Exact Test instead.

Conducting Chi-Square and Log-Linear Analysis in SPSS

  • SPSS Procedure for Chi-Square:

    • Enter categorical variables for training methods and outcomes under weight cases.

    • Utilize crosstab function to specify layout and options.

    • Output includes significant Pearson Chi-Square results and contingency tables with expected counts.

  • Reporting Findings:

    • Include chi-square statistic, degrees of freedom, p-value, and odds ratio in interpretation.

Odds Ratio Calculation

  • Odds of Different Outcomes:

    • Calculate probability ratios for cats based on reward types (food vs affection).

    • Compare odds for both methods and provide numerical results in analysis.

  • Interpretation may look like this:

    • "The odds of a cat dancing in response to food is 6.65 times higher than in response to affection."

Wrap-Up of Analysis Approach

  • For categorical data:

    • The main analytic tools are the Chi-square test (for two variables) and log-linear analysis (for three or more variables).

    • The process involves fitting models, evaluating deviations, assessing interactions, and using odds ratios as effect sizes to interpret results.

  • Students will practice these methods in upcoming labs, ensuring they can report results meaningfully with a focus on significance and effect size.