PSYC 2001 Lecture Notes - Lecture 20: Longitudinal Study, Dependent And Independent Variables, Scatter Plot
PSYC2001 Lec 20
Re-Cap Non-Experimental Design
§ Correlational
§ Two continuous variables
§ Both measured at interval or ratio type (NOT categorical or ordinal)
§ Scatterplot – visually = positive, negative or zero
§ Statistical - is the trend/association different from zero and is the difference
bigger than chance?
§Correlation is represented by r
Issues that Impact Correlations
§ Outliers
§ Extreme scores or combinations of scores
§ Pull esults toads the
§ More problematic with small samples
§ Ex. Someone making 10,000$ a year at 35, vs 19
§ Restriction of range
§ Lack of variability in responses
§ Results from a correlation can be incorrect or misleading when the measure has
restriction of range issues.
§ Ceiling effect: Mainly high scores
§ Floor effect: Mainly low scores
Recap—Experimental Designs
§ We discussed our simplest experimental design
§ 2 variables
§ 1 independent (grouping variable)
§ 1 dependent variable (outcome)
§ “tatistial uestio is ae the eas aeages between the groups different
more than can be expected by chance?
§ When only 2 groups = t-test
§ When 3 or more groups = ANOVA
Inferential Statistics
Statistical Tests
§ Provide a statistical result (r, t, F) and a p-value
§ Remember:
§ We are conducting analyses on our sample
§ Our question is about the population….
find more resources at oneclass.com
find more resources at oneclass.com
§ This is called inference (using data from the sample to infer something
about the population) and thus the term Inferential Statistics
§ The p-value is the probability of getting the result by chance.
§ When p < .05 (less than), small chance that we got result by hae…. Therefore,
olude sigifiat i.e., we found something)
§ Whe p ≥ .5 geate tha o eual to, thee is a high (reasonable) chance that
the result is by chance (fluke)
§ …. Theefoe, olude o-sigifiat (i.e., we found nothing)
§ The p-value is (very) connected with the size of the sample
§ Type I error
§ We conclude there is an effect in our population (i.e. significant result) when
there is NOT
§ Type II error
§ We conclude there is NO effect in our population (i.e. non-significant result)
when there is
Importance
§ Statistical significance
§ Effect size
§ Importance
§just because a finding is significant/non-significant does not mean it is or is not
important.
§ Fo eaple… fidig that the commitment level to a treatment for anxiety results in an
improvement
§ Result r = .04, p = .02, this size of a change is unlikely to be noticeable in a
peso’s life
§ Eaple … fidig that a ediatio does ot hage the life epeta fo a rare
type of cancer
§ Result r = .26, p = .32, non-significant – but there is an effect (not enough people
to hae sigifiat…. “aig lies is ipotat
Really Bad Experiments
find more resources at oneclass.com
find more resources at oneclass.com
Document Summary
Both measured at interval or ratio type (not categorical or ordinal) Scatterplot visually = positive, negative or zero. Someone making 10,000$ a year at 35, vs 19. Results from a correlation can be incorrect or misleading when the measure has restriction of range issues. Tatisti(cid:272)al (cid:395)uestio(cid:374) is (cid:862)a(cid:396)e the (cid:373)ea(cid:374)s (cid:894)a(cid:448)e(cid:396)ages(cid:895) between the groups different more than can be expected by chance? (cid:863) When 3 or more groups = anova. Provide a statistical result (r, t, f) and a p-value. We are conducting analyses on our sample. This is called inference (using data from the sample to infer something about the population) and thus the term inferential statistics. The p-value is the probability of getting the result by chance. When p < . 05 (less than), small chance that we got result by (cid:272)ha(cid:374)(cid:272)e . Whe(cid:374) p . (cid:1004)5 (cid:894)g(cid:396)eate(cid:396) tha(cid:374) o(cid:396) e(cid:395)ual to(cid:895), the(cid:396)e is a high (reasonable) chance that the result is by chance (fluke)