Single factor design Week 5 SRM I Special Session Thu 8 pm, TR202, entirely optional I am not going to re-teach the topics; rather it will be a Q&A format Come with specific questions. You need to know what you dont know (meta-cognition of learning) If you dont know what you dont know, how would I know what you dont know? Google is your best stats tutor There
has rarely been a case, even as a professor, where I have not found an answer to my statistics using Google Single factor: Some basics Factor = independent variable. Note: Factor levels Example: One factor, two levels
Gender (M vs. F) Direction (L vs. R) Example One factor, multiple levels Drug Type (Drug A, Drug B, placebo) Nationality(Indian, Dutch, British) Month (Jan, Apr, Jul, Oct) It is important to know the distinction between factors and levels
because we will be studying factorial designs next week Research design has implications on analyses Research Between-subjects Within-subjects Two levels: independent samples t-test
Two levels: dependent samples t-test > Two levels: one way ANOVA > Two levels: repeated measures ANOVA New in SRM II Between-subjects: Definitions Also called independent groups design, independent samples design
Each group of subjects participates in only one condition of the independent variable. Between-subject variables Between subjects variables are often naturally occurring subject variables Introverted vs. extroverted; Young vs. old; Clinical vs. nonclinical Sometimes
an experimental manipulation artificially creates a subject variable IV: Judges saw a well-dressed or shabbily-dressed accused. DV: How culpable is the accused of the crime? Within-subjects: Definitions Also called dependent groups design, dependent samples design, repeated measures, within-subjects design
All subjects participates in all conditions condition of the independent variable. Analysis of two-level designs Between-subjects Independent samples t-test (you can do the same thing in regression too, but you need to use dummy coding) Within-subjects Dependent samples t-test
Analysis of multilevel designs In SRM I, you learned about one-way ANOVA (one factor, multiple levels, between). One-way ANOVA is only for between-subjects designs But the same concept can be applied to withinsubjects design:
There is no such thing as a one-way ANOVA for repeated measures The concept exist; we just dont have a name for it. SPSS and R simply calls it repeated measures Advantages of each design Betweensubjects Each participant enters the study fresh and nave with respect to the procedures to be tested Easier statistical analyses and interpretation (compared
to dependent groups design) no need to worry about order effects Within-subjects Fewer participants needed For a medium effect size (d = 0.5), you need n = 37 per group to achieve 80% power Can be more meaningful
sometimes E.g., if youre studying change over time, withinsubjects design is the only possible design Disadvantages of each design Betweensubjects Requires larger sample size for hypothesis testing Attrition rates may be
higher in one group than another For a medium effect size (d = 0.5), you need n = 50 per group to achieve 80% power. Wheatgrass Juice group vs. Orange Juice group Differences between groups could be due to other factors unrelated to the manipulation
Within-subjects Complicated counterbalancing may be needed to control for order effects Not possible with every design E.g., you cant manipulate someone to be young and old later to study effects of age on
memory More complex designs You could have X number of between-subjects factors and Y number of within subjects factors. For example: Pre vs. post effects of drugs across gender This is called a mixed design. We will cover this more in depth in Week 7 on Factorial
designs: Mixed designs More complex control groups in research designs Control groups Recall: Control groups are not always needed in experiments, but having them helps provide a useful point of reference A control group can simply be a condition that is different from the experimental group.
Here we look at three controls, used typically in applied psychology. Control groups Placebo control groups Waiting list control groups Yoked control groups (rarely used) Placebo
An effective way to eliminate participant bias (expectancy effects) Participants are given the same treatment that they think will work. Example: Gastric freezing Peter et al. (1962). Technique of gastric freezing in the treatment of duodenal ulcer. J Am Med Assoc. What is gastric freezing? Patients
would swallow a deflated balloon with tubes, and a cold liquid would be pumped for an hour to cool the stomach and reduce acid production, thus relieving ulcer pain. What would a placebo control condition look like? Experimental supercooled fluid Placebo control
? A randomized control trial Gastric freezing: 28 of 82 improved Control group: 30 of 78 improved. z-test of proportions (z = -.57; p = .69): p1 p 2 0.341 0.385 0.044
z 0.568 1 1 1 1 0.231* 0.025 p (1 p ) 0.363* 0.637 82 78 n1 n 2 Ruffin et al. (1962). A cooperative double-blind evaluation of gastric freezing in the treatment of duodenal ulcer. New Eng J Med Never underestimate placebo effects Try this Some
people believe that paying people to pray for them will improve their health How can you devise an experiment with an appropriate control group to test this hypothesis? Waitlist controls Control groups receive no treatment But sometimes, it is unethical to deny a
potentially effective treatment In this case, members of control groups are placed on a waitlist; they will receive the treatment at a later time Waitlist controls Exp Waitlist T1 T2 T3
T4 T5 T6 x x x o o o
DV1 DV2 DV3 DV4 DV5 DV5 o o o
x x X DV1 DV2 DV3 DV4 DV5 DV5
T = Time period X = treatment administered O = treatment not administered DV = dependent measurement obtained (e.g., depression score) Problems with waitlist control How would anyone know if a treatment is potentially effective? What if the experimental treatment takes years? Advanced Issues
Advanced issues: Nested models Researchers want to know whether Textbook A or B results in better student learning. They randomly assigned Class IV classrooms to either use Textbook A or B. Whats the problem? Students in each classroom experience the same teachers, same friends, etc. The data is inherently dependent within each classroom
Advanced issues: Nested models School Class 1 (Book A) Student A Student B Student C Class 2 (Book B) Student D Student E Student F Advanced issues: Nested models Clinicians want to know whether group therapy
is better than waitlist control for treating anxiety. They randomly assigned patients to either group therapy or waitlist control group. Whats the problem? Partially nested data: patients in group therapy have dependent scores, whereas patients in waitlist control have independent scores Advanced issues: Partially nested models Group treatment
Each patient score is dependent on who their fellow patient is within the group treatment Waitlist control Each patients score is independent Discussion We discussed classrooms and group therapy sessions as examples where data is inherently nested.
What other situations can you think of where the data is inherently nested? Even more complex nested models Big data is the latest trend Solutions We will not learn how to deal with nested data in SRM II; this is a graduate topic But you need to learn to recognize it
Curious students can consult advanced statistical textbooks on hierarchical modeling or multilevel modeling Take home messages Your design determines your statistical analyses Good designs help reduce the number of alternative explanations
Skills of the future: The ability to design and analyze complex data structures may be the future. Pick up R and Python if you want to be in it.
Books Paperbacks like Diary of a Wimpy Kid Hardcover fiction like The Mazerunner and Legend Nooks and e-books via Follett Shelf Audiobooks on playaway, CD, MP3 CD, cassettes and downloadable via Catalist Digital Nonfiction on every subject from the Abominable...
Chisholm v. Georgia (1793) - lawsuit involving a state being sued by a citizen from another state could be heard in federal court. concerns over the extent of federal power - amendment limits jurisdiction of federal courts in cases of...
(Williams v Fawcett) Extra exception for criminal cases - "in the interests of justice" Taylor or Gould How has precedent been used to develop the law Law reporting Law of Contract (Carlill vThe Carbolic Smoke Ball Co) Law of Tort...
2) In solution: Zno and Cu2+ collide and electrons are transferred No work can be obtained; only heat is generated 3) In separate compartments, electrons must go through a wire = Galvanic Cell a) Generates a current = moving electrons...
Insurers are taking Risk Management seriously Thanks to Garth Symington & Rob Perkins. Their interaction with all areas of tanker industry is the main driver in us working together. Put your logo here on master template (optionally) Put your logo...
Broken links kept to a minimum NASA web policies/registration & section 508 Improvements planned for search facility Comments Welcome Please comment (now or later) on Areas needing improvement Areas to be removed Areas to be expanded Navigation improvement suggestions Font/color...