Data Lab # 4 June 16, 2008 Ivan Katchanovski, Ph.D. POL 242Y-Y

Data Lab # 4 June 16, 2008 Ivan Katchanovski, Ph.D. POL 242Y-Y

Chi-Square Test: Research and Null Hypotheses, Sample, and Weights Research hypothesis: Men in Canada are more confident in television than women are The null hypothesis: Men and women in Canada have the same confidence in television Canadian sample of the 2000 World Values Survey (WVS) – Select Canadian respondents from the 2000 WVS dataset Identify value for Canada (12) in “nation” variable (v2) Data-Select cases-If: v2 12 – Select Weight: Data-Weight Cases-Weight Cases By: v245 2

Chi-Square Test: Dependent and Independent Variables Dependent variable – Confidence in Television (v150) Check if recoding values and defining missing values needed – Define -4, 8, 9 as missing values Independent variable: – Sex (v223) Check if recoding values and defining missing values needed – Define -4, 8, 9 as missing values 3

SPSS Commands SPSS Command: – Analyze Descriptive Statistics-Crosstabs – “Row” box: select dependent variable (Confidence in Television) – “Column” box: select independent variable (Sex) – “Cells” Option: Column percentages – “Statistics” Option: Chi-square SPSS Output – Check % of cells that have expected count less than 5 If large %, collapse categories by recoding values 4

Results SPSS Output: – Pearson Chi-square: Asymp. Significance p .317 p .05 Statistically insignificant Accept the null hypothesis Reject the research hypothesis: – Canadian male and female respondents do not differ significantly in their confidence in television – The gender differences are statistically insignificant 5

Modified Hypothesis New research hypothesis: Confidence in television in Canada differs significantly by the level of education The null hypothesis: Confidence in television in Canada is independent of the level of education Select new independent variable: – Education (v226) – Check if recoding values and defining missing values needed Define -4, 98, 99 as missing values 6

Chi-Square Test: SPSS Command SPSS Command: – Analyze Descriptive Statistics-Crosstabs – “Row” box: select dependent variable (Confidence in Television) – “Column” box: select independent variable (Education v226) – “Cells” Option: Column percentages – “Statistics” Option: Chi-square SPSS Output – Check % of cells that have expected count less than 5 16.7% of cells that have expected count less than 5 – Need to collapse categories 7

Collapsing/Recoding the Independent Variable Collapse /recode v226 variable into new variable SPSS recoding command – Transform/Recode into Different Variables – SPSS “Old and New Values”: Range from 1 to 7 0 Range from 8 to 9 1 Label values of the recoded education variable (renamed as university education): – 0“Less than university” – 1 “University” – Dummy variable: two values (0 and 1) 8

Results SPSS Output: – Pearson Chi-square: Asymp. Significance p .000 p .001 Statistically significant Reject the null hypothesis Accept the research hypothesis: – Confidence in television in Canada differs significantly by the level of education – The differences are statistically significant at the .001 or .1% level – 35% of the respondents without university education, compared to 27% of the respondents with university education, have quite a lot of confidence in television 9

Presenting Results Table 1. Confidence in television in Canada by education level, 2000 World Values Survey, % Less than university University A great deal 6 3 Quite a lot 35 27 Not very much 46 55 None at all 13 15 Total, % 100 100 N 1363 524 10

Independent Samples t Test Hypothesis testing: Comparison of the sample means of a dependent variable for two groups that differ on an independent variable Dependent variable: – interval-ratio – ordinal variable treated as interval ratio Research Hypothesis: Men have lower level of confidence in television compared to women in Japan Null hypothesis: Men and women in Japan have the same level of confidence in television 11

Independent Samples T-test: Dependent and Independent Variables Dependent variable: Confidence in television (v150) Ordinal but can be treated as interval-ratio: – If defined as a measure of non-confidence in television because increase in its values means decrease in confidence in television – If recoded by assigning higher values to higher confidence levels Independent variable: Sex Two categories (groups): Male and Female – Sampling distribution: Student t distribution – SPSS selects t distribution automatically 12

Independent Samples T-test: SPSS Commands Select Japan sample in the 2000 WVS dataset – Identify value for Japan (13) in “nation” variable (v2) Data-Select cases-If: v2 13 – Check if weights should be used SPSS Command: – – – – Analyze-Compare Means-Independent Samples T Test Select Confidence in television (v150) as “Test Variable” Select Sex as “Grouping Variable” Define Groups: “Use specified values”: 1 (Male) and 2 (Female) 13

Results Compare means of the dependent variable – Men have higher mean level of non-confidence in television compared to women in Japan: 2.33 2.19 Select “equal variances assumed” if Leven’s test for equality of variances is statistically insignificant: – Sig. p .000 .05 Select “equal variances not assumed” if Leven’s test for equality of variances is statistically significant: – Sig. p .05 Determine significance in t-test for equality of means – p .000 p .001 Statistically significant 14

Interpretation of Results Reject the null hypothesis Accept the research hypothesis: – Men have lower level of confidence in television compared to women in Japan – The differences are statistically significant at the .001 or .1% level 15