Transcript Data Analysis

Soc 3306a Lecture 7:
Inference and
Hypothesis Testing
T-tests and ANOVA
Hypothesis Tests
When sample is non-random or have nonnormal distribution, use Chi-square test
 Non-parametric (can not generalize)
 More powerful method is inferential test
 Through the use of parametric statistics
 Assumptions: random sampling, normal
distribution and relatively equal variances

Inference
When assumptions met, can use sample
statistics to generalize to population
 Test Ho (null hypothesis) of “no difference”
 Find evidence for H1 (alternate or
research hypothesis)
 Caution: when using very large samples,
even trivial differences become significant
 Always check actual mean differences too

Alpha Levels






N<1000, alpha = .05
N>1000, alpha = .01 or .001
Your p-value is the probability associated with
the statistic you used
Smaller p-value = stronger evidence for your
research hypothesis
Eg. p-value of <.001 means that you would find
that result less than 1 in 1000 times
Strong evidence for research hypothesis in
population
Single Sample T-Test
Figure 1

For use when population value is known
 This
is entered as the “test value”
Is the sample significantly different form
the population?
 Can use to test differences in means
 Requires a DV at interval-ratio level data
 For nominal level (%) need to use the
binomial test (non-parametric)

Independent Samples T-Test
Figure 2
For testing differences in two sample means
 DV = interval-ratio is entered as “test
variable”
 IV is your “grouping variable” – binary
 Can be nominal or ordinal
 Need to “define groups” (enter the codes for
the categories (2 groups) to be tested)
 Also look at confidence interval of difference

Oneway ANOVA (F-test)
Figure 3
To test for differences in 3 or more means
 DV is I-R and IV is nominal/ordinal level
 Assumptions: relatively equal variances
and group sizes but F is fairly “robust”
 Levene statistic to test for equal variances
 Post Hoc tests

 Bonferroni:
confidence intervals of differences
 Tukey B: to examine means

Can also ask for “Means Plots” - graph
Univariate Analysis of Variance
Figure 4
Like oneway ANOVA but more flexible and
informative (see Babbie Ch. 14 for detail)
 Can use for 1 or more IV’s at a time
 Tests “main effects” and when used for 2+
IV’s, tests “interaction effects” (“Two-way”)
 Produces regression-like output and can
also be combined with regression to
examine coefficients and plots (“ANCOVA”)
