independent samples t test

0.0(0)
Studied by 0 people
call kaiCall Kai
learnLearn
examPractice Test
spaced repetitionSpaced Repetition
heart puzzleMatch
flashcardsFlashcards
GameKnowt Play
Card Sorting

1/44

encourage image

There's no tags or description

Looks like no tags are added yet.

45 Terms

1
New cards

Between-Group Research Design (Between-Subjects Design)

A research design that compares two or more groups of participants where each condition is comprised of completely different participants and each individual contributes only one score. Discussed in lecture with experimental examples (randomly assigning participants to treatment vs. control groups, online vs. in-person classes, or self-focused vs. other-focused COVID-19 messages) and non-experimental examples (splitting participants by pre-existing characteristics such as young vs. old, grizzly bears vs. humans, or psychology vs. sociology majors with no double major). The textbook defines independent samples as having no special relationship between observations across the two groups, such as students in Anastasia's vs. Bernadette's tutorials. (Slides; Transcript 11.12 56:44-59:54, 11.17 05:45-06:40; Textbook pp. 248-249)

2
New cards

Independent-Samples t-Test

An inferential statistical test appropriate for between-group designs with exactly two groups where each individual contributes one score, testing whether the two populations share the same mean by evaluating the sample mean difference between the two groups. Illustrated in lecture using Luttrell and Petty's pandemic messaging study (Other-Focused n = 99, M = 3.91, SD = 0.927 vs. Self-Focused n = 92, M = 3.74, SD = 1.00; t(189) = 1.22, p = .225) and in the textbook using Dr. Harpo's class grades (Anastasia n = 15, M = 74.53, SD = 9.00 vs. Bernadette n = 18, M = 69.06, SD = 5.77; t(31) = 2.12, p = .04253). (Slides; Transcript 11.12 59:54, 11.17 05:33-06:40; Textbook pp. 248-253)

3
New cards

Null Hypothesis (Independent-Samples t-Test)

A population-level hypothesis targeting the mean difference between two groups and typically specifying that nothing is going on or no difference exists: H0: mu1 - mu2 = 0 (or mu1 = mu2). In lecture, the null hypothesis states that self-focused and other-focused health messages are equally persuasive at the population level (visualized as two group distributions sharing the same population mean); in the textbook, it states that Anastasia's and Bernadette's students have identical population mean grades. (Slides; Transcript 11.17 07:13-08:59, 10:17-10:33; Textbook pp. 249-250)

4
New cards

Alternative Hypothesis (One-Tailed vs. Two-Tailed)

The population-level hypothesis reflecting the researcher's interest or belief that a group difference exists. A two-tailed (non-directional) alternative specifies that the two population means differ in either direction (Ha: mu1 - mu2 != 0 or mu1 != mu2, e.g., self-focused and other-focused messages are not equally persuasive), whereas a pre-planned one-tailed (directional) alternative specifies that one particular group is higher than the other (Ha: mu1 > mu2 or Ha: mu1 < mu2, e.g., predicting other-focused is more persuasive than self-focused). (Slides; Transcript 11.17 09:05-10:33, 59:15-1:01:51; Textbook p. 249)

5
New cards

Pooled Standard Deviation and Pooled Variance

A single combined estimate of variability used in Student's independent-samples t-test under the assumption that both populations have identical standard deviations (sigma1 = sigma2). Calculated as the square root of a weighted average of the two sample variances weighted by each group's degrees of freedom (n1 - 1 and n2 - 1) divided by total degrees of freedom (n1 + n2 - 2): s_p = sqrt(((n1 - 1)s1^2 + (n2 - 1)s2^2) / (n1 + n2 - 2)). The textbook also expresses pooled variance as the sum of squared deviations from each observation's own group mean divided by N - 2. (Slides; Transcript 11.17 16:52-17:56, 1:11:12-1:11:45; Textbook pp. 249-252, 266)

6
New cards

Standard Error of the Difference Between Means

The standard deviation of the sampling distribution of group mean differences across many repeated two-group studies of equal sample sizes (n1 and n2), representing the average or expected discrepancy between a sample group mean difference and the true population mean difference. Computed in Student's test as s_{xbar1 - xbar2} = s_p * sqrt(1/n1 + 1/n2); in lecture, the pandemic dataset yielded s_{xbar1 - xbar2} = 0.140 points, meaning two random samples of sizes 99 and 92 should on average differ from 0 by 0.140 points if H0 is true (compared to SE = 2.59 in Dr. Harpo's data). (Slides; Transcript 11.17 14:00-25:14; Textbook pp. 252-253)

7
New cards

Sampling Distribution of Group Mean Differences

A theoretical distribution built by imagining 10,000 independent researchers each collecting two samples of the exact same sizes (e.g., n1 = 99, n2 = 92), calculating each study's group mean difference (xbar1 - xbar2), and plotting all 10,000 mean differences. By the Central Limit Theorem, this distribution approximates a normal curve centered on the true population mean difference (anchored at 0 under H0) with spread equal to the standard error. (Slides; Transcript 11.17 15:09-25:49; Textbook p. 252)

8
New cards

Independent-Samples t-Statistic

A standardized test statistic converting the difference between the sample estimate (xbar1 - xbar2) and the hypothesized population mean difference (mu1 - mu2) into standard error units: t = ((xbar1 - xbar2) - (mu1 - mu2)) / s_{xbar1 - xbar2}. In lecture, the pandemic study yielded t = (3.91 - 3.74 - 0) / 0.140 = 1.22, meaning the observed mean difference is 1.22 standard error units higher than 0 (1.22 times as large as expected due to sampling error alone); in the textbook, Dr. Harpo's classes yielded t = 5.48 / 2.59 = 2.12. (Slides; Transcript 11.17 26:32-32:08; Textbook pp. 249, 252-253)

9
New cards

Degrees of Freedom and Critical Value (Independent-Samples t-Test)

Degrees of freedom for Student's independent-samples t-test equal n1 + n2 - 2 (N total observations minus 2 sample-mean constraints). In lecture, df = 99 + 92 - 2 = 189 produced two-tailed critical values of +/-1.973 in jamovi, meaning 95% of samples under H0 fall between -1.973 and +1.973; because observed t = 1.22 fell inside these bounds, the result was non-significant (fail to reject H0). In the textbook, df = 15 + 18 - 2 = 31 yielded t(31) = 2.12, p = .04253. (Slides; Transcript 11.17 35:20-40:44; Textbook pp. 252-253)

10
New cards

p-Value (Independent-Samples t-Test)

The conditional probability, assuming the null hypothesis of zero population difference is true, of drawing a sample with a t-statistic at least as extreme as the observed value. In lecture, two-tailed p = .225 indicated a 22.5% probability of observing t >= +1.22 (.113) or t

11
New cards

95% Confidence Interval for Population Mean Difference

An interval computed as Estimate +/- (Critical Value * Standard Error) estimating the plausible range for the true population mean difference (mu1 - mu2). In lecture, 0.170 +/- (1.973 * 0.140) yielded [-0.106, 0.446] (ranging from self-focused higher by .106 to other-focused higher by .446); because 0 is inside this interval, H0 cannot be rejected. In the textbook, Dr. Harpo's 95% CI [0.20, 10.76] excludes 0, confirming statistical significance. (Slides; Transcript 11.17 50:13-56:41; Textbook pp. 252-253)

12
New cards

Cohen's d Effect Size (Standardized Mean Difference)

A metric of practical significance independent of sample size that expresses the group mean difference in pooled standard deviation units (placing it on a z-score metric): d = (xbar1 - xbar2) / s_p. In lecture, the pandemic study yielded d = (3.91 - 3.74) / sqrt(((99-1)0.927^2 + (92-1)1^2)/(99+92-2)) = 0.176, falling in the negligible magnitude range (< |.20|); in the textbook, Anastasia vs. Bernadette yielded d = 0.74 (sometimes called Hedges' g when using pooled SD). (Slides; Transcript 11.17 1:09:19-1:14:42; Textbook pp. 266-267)

13
New cards

jamovi Independent Samples T-Test Workflow

Running Analyses -> T-Tests -> Independent Samples T-Test, moving the continuous outcome variable (e.g., pers or grade) into Dependent Variables and the categorical two-group factor (e.g., message.F or tutor) into Grouping Variable, keeping Student's checked under Tests, selecting the two-tailed or one-tailed option under Hypothesis, and checking Mean difference, Confidence interval, Effect size, and Descriptives (plus Normality and Equality of variances under Assumption Checks). (Slides; Transcript 11.17 12:06; Textbook pp. 252-253)

14
New cards

APA-Style Summary (Independent-Samples t-Test)

A standardized four-part summary stating (1) the test used and research question, (2) a reference to Table 1 containing group M, SD, and n, (3) the statistical significance result with t(df), p-value, and 95% CI for the mean difference (e.g., t(189) = 1.22, p = .225, 95% CI [-.106, .446] in lecture; t(31) = 2.1, p < .05, CI95 = [0.2, 10.8] in textbook), and (4) Cohen's d alongside its practical magnitude benchmark (d = .176 negligible in lecture; d = .74 in textbook). (Slides; Textbook p. 253)

15
New cards

Dr. Harpo's Class Dataset (harpo.csv)

A textbook dataset of 33 students (variables: ID, grade, tutor) comparing Anastasia's tutorial (N = 15, M = 74.53, SD = 9.00, SE = 2.32) against Bernadette's tutorial (N = 18, M = 69.06, SD = 5.77, SE = 1.36) when overall class grading rules are unknown. (Textbook pp. 248, 253)

16
New cards

Sign of the t-Statistic and Group Ordering

Because jamovi designates Group 1 and Group 2 arbitrarily, reversing which group is subtracted first flips the signs of the t-statistic and confidence interval bounds while keeping the two-sided p-value identical; researchers should report the t sign to match their prose phrasing (positive for "higher", negative for "lower"), a trick that applies only to t-tests and not chi-square or F-tests. (Textbook pp. 254-255)

17
New cards

Assumptions of Student's Independent-Samples t-Test

Three assumptions underlying Student's test: (1) Normality (both groups are drawn from normally distributed populations), (2) Independence (observations are independent both within each group and across groups with no shared participants), and (3) Homogeneity of variance or homoscedasticity (both populations have identical standard deviations). (Textbook p. 255)

18
New cards

Levene's Test for Equality of Variances

An assumption check in jamovi's Independent Samples T-Test evaluating whether the two groups have equal population variances; a low p-value suggests the equal-variance assumption is violated (in harpo.csv, Levene's F = 2.49, df = 1, p = 0.12508, showing no violation). (Textbook pp. 253, 255)

19
New cards

Shapiro-Wilk Normality Test in Independent-Samples t-Test

An assumption check accessed via Assumption Checks -> Normality in jamovi producing a W statistic and p-value to evaluate whether data depart from normality (in harpo.csv, W = 0.98, p = 0.82663, indicating no violation). (Textbook p. 253)

20
New cards

Welch's Independent-Samples t-Test (Welch 1947)

An alternative version of the independent-samples t-test (selectable in jamovi) that drops the restrictive homogeneity of variance assumption, allowing the two populations to have different standard deviations under the alternative hypothesis while still assuming normality and independence. (Textbook pp. 248, 255-256)

21
New cards

Welch Standard Error and Fractional Degrees of Freedom

Because pooling unequal standard deviations is "averaging apples and oranges", Welch's t-test calculates standard error directly from each group's unpooled variance and sample size as SE(Xbar1 - Xbar2) = sqrt(s1^2/N1 + s2^2/N2), and calculates degrees of freedom in a different way that does not have to be a whole number. (Textbook p. 256)

22
New cards

Hedges' g Statistic (Hedges 1981)

An alternative name sometimes applied to the standard Student independent-samples Cohen's d formula that divides the sample mean difference by the unbiased pooled standard deviation estimate (sigma_hat_p). (Textbook p. 266)

23
New cards

Glass' Delta

A variation of standardized mean difference effect size that uses the standard deviation of only one of the two groups (such as a control group) in the denominator when that group is a purer reflection of natural variation than the treatment group. (Textbook p. 266)

24
New cards

Uncorrected Sample Effect Size (Dividing by N)

A variant of Cohen's d that omits the N - 2 bias correction in the pooled standard deviation denominator and divides by N instead, appropriate only when describing sample effect size rather than estimating population effect size. (Textbook p. 266)

25
New cards

Hedges and Olkin (1985) Effect Size Bias Correction

A correction formula that adjusts for slight bias in the usual pooled Cohen's d estimator by multiplying the calculated d value by (N - 3) / (N - 2.25). (Textbook p. 266)

26
New cards

Experimental vs. Non-Experimental Between-Group Applications

Slides contrast Experimental applications (randomly assigning participants to treatment vs. control conditions, such as the pandemic health message groups) with Non-Experimental applications (splitting participants based on pre-existing characteristics, such as young vs. old, grizzly bear vs. human, or psychology vs. sociology majors). (Slides; Transcript 11.12 57:03-59:54, 11.17 05:54-06:33)

27
New cards

Evaluations of COVID-19 Social Distancing Messages (Luttrell & Petty)

The primary lecture dataset comparing persuasiveness (pers) of an "Other-Focused" message ("protect your community", n = 99, M = 3.91, SD = 0.927, SE = 0.0931) and a "Self-Focused" message ("protect yourself", n = 92, M = 3.74, SD = 1.00, SE = 0.105), yielding Mean difference = 0.170, SE difference = 0.140, t(189) = 1.22, two-tailed p = .225, 95% CI [-0.106, 0.446], and Cohen's d = 0.176. (Slides; Transcript 11.17 05:45-06:33, 12:22-12:56)

28
New cards

Three Equivalent Ways to Test the Null Hypothesis

Making the reject vs. fail-to-reject decision via (1) comparing the t-statistic to the critical value (+/-1.973), (2) comparing the p-value to .05, or (3) checking whether 0 falls inside the 95% confidence interval; all three methods always reach the exact same conclusion. (Slides; Transcript 11.17 34:37-35:13, 50:21)

29
New cards

Comparison of One-Sample, Paired-Samples, and Independent-Samples t-Tests

A synthesis taught in slides and lecture distinguishing the three t-tests by research design: One-sample tests whether a known/hypothesized population mean is supported by one sample mean (e.g., Trader Joe's orange sweetness = 5, or psychology majors averaging 7 hours of sleep); Paired-samples tests two conditions where each individual or dyad contributes 2 scores (e.g., half orange fresh vs. half frozen, or sleep before vs. after midterm); Independent-samples tests two conditions where separate individuals contribute 1 score each (e.g., separate oranges fresh vs. frozen, or psychology vs. sociology majors' sleep with no double major). (Slides; Transcript 11.12 1:05:17-1:12:39)

30
New cards

Confidence Intervals Across the Three Types of t-Tests

Distinguishing what the 95% confidence interval estimates in each test: the possible range of the single population mean (one-sample t-test), the possible range of the population mean difference/change score within one population (paired-samples t-test), and the possible range of the population mean difference between two independent conditions/groups (independent-samples t-test). (Slides; Transcript 11.17 56:46-58:39)

31
New cards

Pros and Cons of Within-Group vs. Between-Group Designs

Within-group designs allow participants to serve as their own control and provide higher statistical power, but risk carryover effects from prior conditions that are hard to wash out; between-group designs avoid carryover effects and are required when conditions cannot overlap (e.g., young vs. old, grizzly bears at 35 mph vs. female humans at 6.5 mph), though each participant provides only one score. (Slides; Transcript 11.12 58:42-1:03:36)

32
New cards

Elden Ring vs. Animal Crossing Design Example

A lecture illustration showing how research design determines the test for the same question (whether Elden Ring causing motion sickness/arrhythmia vs. Animal Crossing calming players affects emotional stability): having the same players try Animal Crossing first and Elden Ring second gives 2 scores per person (paired-samples t-test), whereas assigning separate players to play only Animal Crossing for 10 weeks or only Elden Ring for 10 weeks gives 1 score per person (independent-samples t-test). (Slides; Transcript 11.12 1:03:40-1:05:17)

33
New cards

Cohen's d Effect Size Guidelines (Slides)

Specific cutoffs presented on slides and in lecture for interpreting absolute Cohen's d magnitude: Negligible (< |.20|), Small (|.20 to .50|), Moderate (|.50 to .80|), and Large (> |.80|). (Slides; Transcript 11.17 1:12:59-1:13:21)

34
New cards

Statistical Significance vs. Practical Significance

Distinguishing whether data refutes H0 (statistical significance via t, p, and CI, which is sensitive to sample size because larger N shrinks SE = s/sqrt(N) and inflates t) from whether an effect is meaningfully large in real life (practical significance via Cohen's d, which divides by SD and is independent of N). Illustrated by paying $10,000 for a statistically significant 1-point or 0.5-point SAT improvement vs. a 500-point improvement. (Slides; Transcript 11.17 1:05:52-1:10:27)

35
New cards

p-Hacking via Sample Size Inflation or Switching Tails

Professor Du's warning against two unethical practices: continuously recruiting extra participants (e.g., adding 10 at a time) until SE shrinks enough to produce p < .05, or changing from a two-tailed to a one-tailed hypothesis after inspecting the p-value; prevented by pre-registering hypotheses, sample size, and analysis plans prior to data collection. (Transcript 11.12 05:05-05:27, 11.17 10:41-11:30, 1:07:38-1:08:27)

36
New cards

distrACTION Module Critical Value Lookup

Computing exact two-tailed critical values in jamovi's T-Distribution tool by entering df = n1 + n2 - 2 (e.g., df = 189, lambda = 0), selecting Compute quantile(s) with p = 0.95 and central interval quantiles, yielding x1 = -1.973 and x2 = 1.973. (Slides; Transcript 11.17 36:10-36:25)

37
New cards

Conceptual Unity of Standard Error Across Tests

Professor Du's exam strategy explaining that standard error is always the standard deviation of the sampling distribution (average distance between sample estimate and population parameter), differing across tests only in the component plotted in each sample: sample means (one-sample), mean change scores (paired-samples), or group mean differences (independent-samples). (Transcript 11.17 40:50-43:27)

38
New cards

Can a paired-samples t-test and an independent-samples t-test answer the same research question, and what determines which one to use?

Yes, both can answer the same research question (e.g., whether freezing oranges alters sweetness, or whether online vs. in-person classes differ), and the research design determines which test to use: use paired-samples when each individual or dyad contributes two linked scores (within-group design), and independent-samples when separate participants in two groups contribute one score each (between-group design). (Slides; Transcript 11.12 56:27-58:34, 1:05:58-1:08:19)

39
New cards

What test must be used for a between-group design with three or more independent groups instead of an independent-samples t-test?

An ANOVA (Analysis of Variance) must be used, because the independent-samples t-test is appropriate only for between-group designs with two groups. (Transcript 11.12 1:00:00-1:00:05)

40
New cards

Why is the degrees of freedom for Student's independent-samples t-test equal to n1 + n2 - 2 rather than N - 1?

Because the dataset contains n1 + n2 total observations and two separate sample means (one for each group) are estimated as constraints, leaving n1 + n2 - 2 degrees of freedom (e.g., 99 + 92 - 2 = 189). (Slides; Transcript 11.17 35:40-36:03; Textbook pp. 251-252)

41
New cards

In the COVID-19 messaging study (t = +1.22, two-tailed p = .225), how does the p-value change under each of the two possible one-tailed alternative hypotheses?

When predicting Other-Focused > Self-Focused (matching the positive observed t = +1.22), only the upper tail at or above +1.22 is counted, halving the two-tailed p-value to p = .113; when predicting Self-Focused > Other-Focused (opposite to the observed data), "at least as extreme" covers everything from +1.22 down into the negative tail, yielding p = 1 - .113 = .887. (Slides; Transcript 11.17 59:15-1:05:06)

42
New cards

What is the mathematical difference between the t-statistic and Cohen's d in an independent-samples t-test, and what units does each use?

Both place the sample group mean difference (xbar1 - xbar2) in the numerator, but the t-statistic divides by the standard error (and uses standard error units, growing larger as sample size N increases), whereas Cohen's d divides by the pooled standard deviation (and uses standard deviation units, remaining independent of sample size). (Slides; Transcript 11.17 1:09:19-1:12:56; Textbook pp. 252, 266)

43
New cards

When observing t = 1.22 within the critical bounds of +/-1.973 and 0 inside the 95% confidence interval [-0.106, 0.446], why do we say "fail to reject the null hypothesis" rather than "accept the null hypothesis"?

Because failing to reject means only that the observed data are plausible under the null hypothesis (H0 remains "innocent"), not that H0 has been proven true—a real population difference might still exist that the sample lacked sufficient evidence to detect. (Slides; Transcript 11.17 40:17-40:44, 49:40-50:08, 58:39-58:54)

44
New cards

Why did Professor Du clarify that "400 students" collected by a researcher must be a sample when standard errors and confidence intervals are calculated?

Because standard error and confidence intervals are inferential statistics used only when you have a sample estimate and need to infer an unknown population parameter; if you already had the entire population, neither standard error nor confidence intervals would be needed. (Transcript 11.17 01:37-03:04)

45
New cards

Why does the independent-samples t-test use a t-distribution rather than a normal distribution?

Because under the Central Limit Theorem the sampling distribution only approximates a normal distribution and relies on estimated sample variances rather than known population parameters, making the heavier-tailed t-distribution more accurate, particularly when sample sizes are small. (Transcript 11.17 20:26-21:00; Textbook p. 252)