Experimental and Correlational research designs Author: Owen Emlen [Email] Copyright © 2006 Owen Emlen, All Rights Reserved These slides were extracted from a series of two PowerPoint presentations created by the author Owen works as an independent consultant and is a member of BrainTech, LLC, where he creates online applications that analyze data that look for trends in behavior. Here is an example model Target Audience The target audience consisted of undergraduate students enrolled in a 300-level research design and statistics course offered through the Psychology Department at Western Washington University The course instructor was Professor David Sattler, who was out of the country doing research for the week The Importance of Research Designs and Methods Statistics cannot salvage a poorly designed experiment! 1. What is wrong with the question below? “I believe that homosexual marriage is wrong and therefore they should not be allowed to adopt children, live together, or hold hands in public.” Strongly Agree - - - - - - - - Strongly Disagree 1 2 3 4 5 A poorly designed experimental manipulation 2. “We showed participants angry or clown faces in an attempt to make them angry or happy:” More Pitfalls: Confounds A researcher is measuring differences in talking speed between men and women 1. Participants are asked to converse in separate waiting areas… 2. One of the waiting areas has a coffee vendor! Women are observed in the morning, and men at night Two types of research designs Experimental designs involve manipulation of one or more Independent Variables (IV) Causality can potentially be inferred Correlational designs examine the relationship between two or more existing (non-manipulated) variables Experimental Design Example study: Level of Arousal and Perceived Attractiveness* Method 1. 2. 3. Participants are asked to meet a confederate (of the opposite sex) in a park at one of two locations: (1) on a garden path, or (2) past the garden path, across a rickety old suspension bridge. Participants were asked some standard demographic questions by the confederate. Participants returned to their car, where they were asked to rate the attractiveness of the confederate. *Hypothetical study based on a prior research by Dutton and Aron (1974) Experimental Design, cont. First, is this an experimental design? What are we measuring / What is the IV, DV? What was the manipulation? (N=40) Low arousal (Met on the garden path) High arousal (Walked across the rickety suspension bridge) Attractiveness rating M=7.1 M=8.4 This is a 1x2 design (rows x columns) Low arousal (Met on the garden path) High arousal (Walked across the rickety suspension bridge) Attractiveness rating by Female participants (n=20) M=7.0 M=7.1 Attractiveness rating by Male participants (n=20) M=7.2 M=9.5 • Now this is a 2x2 design • 1 true independent variable, 1 status (quasi-independent) variable • 1 Dependent Variable (DV): Attractiveness Low arousal (Met on the garden path) Medium arousal High arousal (Passed a barking (Walked across the rickety dog) suspension bridge) Attractiveness rating by Female participants (n=20) M=7.0 M=7.1 M=6.7 Attractiveness rating by Male participants (n=20) M=7.2 M=8.5 M=9.3 Now what? A 2x3 design • Note, there are still 2 IVs and 1 DV • One participant (quasi-independent) variable (Sex) • One manipulated IV (arousal level), with 3 levels • Attractiveness rating is still our DV Adding more “dimensions” to the table A 2x2x2 design would indicate 3 independent (or quasi-independent) variables A 2x2x2 design: 1. 2. 3. You can picture this as a cube, or a 3-D table Quasi-IV (Participant Variable): Sex (male/female) True IV: Arousal level: (low/high - path or bridge) Quasi-IV (Participant variable): Introversion/Extroversion There is still only one DV (attractiveness) Clarification: Participant (or status) Variables What is a participant (or status) variable? Hint: If we recruited participants only from an acrophobia (fear of heights) support group, how aroused might they be when walking across the bridge? If we recruited participants only from a climbing club, how aroused might they be in the same situation? Random selection and Random assignment of participants is important How can we be sure that having participants walk over the bridge had any effect? Manipulation Checks Were participants really more aroused after walking over the bridge? A manipulation check allows a researcher to confirm that his/her manipulation was successful Ideas? Measure pulse and blood pressure Give participants a survey – ask them Correlational Design Correlational research examines the relationship between two or more non manipulated variables. What is the relationship between: 1. 2. 3. 4. Height and weight? Birth order and years of education? Cigarettes smoked per day and health care costs? How close to the front you sit in a classroom and your grade in a class? What can correlational research tell us? Imagine that researchers find an association between sitting in the front of the classroom and receiving good grades You promptly move to the front of the classroom, and expect your grade will improve Don’t bet money on it… Correlation and Causality With correlational research designs, causality cannot be inferred Example: Researchers want to investigate the link between religious affiliation and alcohol consumption* They measure the number of bars and churches in randomly-selected towns # of Bars # of Churches ? * Example by Professor Kristi Lemm of Western Washington University Pitfalls of correlational research designs The researchers find that towns with more bars also have more churches Therefore, religious persons tend to drink more, or perhaps alcohol consumption is a reason people attend church …what is wrong with these conclusions? We cannot infer causation! Larger towns tend to have more bars and more churches. Therefore, a third (and more likely) explanation: # of Bars Town Population # of Churches A more realistic example Example: Researchers find that people who had a psychotic episode tended to report high levels of marijuana use just prior to the episode Does this imply that marijuana causes psychotic episodes? It may, but be careful! It is difficult to establish causality via observation alone Example, cont. 1. There may be a “third” explanation: 2. High levels of stress may be a reason for increased drug use, and high levels of stress are associated with psychotic episodes Which came first, the chicken or the egg? Perhaps some people self-medicate to deal with the early symptoms associated with a psychotic episode Does this imply that the onset of a psychotic episode leads to increased drug use? Trying to guess what causes what gets messy fast! Don’t do it! But what if there really is a cause? How would we confirm causality? Simply By use an experimental design manipulating a variable can we infer causality So, the solution is simple: Take 10,000 people, randomly assign half of them to consume large amounts of marijuana each day, and measure how many … people from each group have a psychotic episode Ethics and practicality Ethics: Should researchers create Tsunamis by detonating sub-oceanic nuclear bombs in order to study the effects of “natural” disasters? Practicality: Randomly assign half of your participants to cut out all caffeine intake for the next year… Critical Reading: Correlational Research Designs Creative interpretation of data is not limited to statisticians, politicians, and lawyers! “Headline: Children who sit in the back of the classroom receive lower grades than those who sit in the front.” Are there implied overgeneralizations? 2. Does the use of present tense affect your interpretation? 3. Would a parent whose child is doing poorly be more or less likely to assume causality? 1. Critical Evaluation, cont. Possibly implied: “[All] Children who sit in the back of the classroom [always] receive worse grades than [each and every child] who sits in the front.” Better: “Researchers X and Y found that children who sat in the back of the classroom were more likely to receive lower grades than those who sat in the front.” Critical Evaluation: Summary Be objective and know your biases; emotion affects interpretation 2. Is there mention of the research design? 3. Look for the Manipulated Variable. If nothing was manipulated, consider carefully before assuming causality 1. For Discussion Headline: “People who swim in the ocean have five times the average level of toxic mercury in their bloodstream!” Time to cancel your Snorkeling Adventure? Swimming example, cont. What groups were the researchers comparing? People who swim in the ocean (how often?) The “average level” of mercury may include people who live in the Midwest who rarely eat seafood Perhaps people who swim in the ocean tend to live near the ocean and eat more fish, and the fish have high levels of mercury Still, this relationship between ocean swimmers and high blood mercury levels contains important information! Critical Evaluation, cont. The data are still extremely valuable and may indeed hint at some underlying cause Research design is cyclic in nature; observation can lead to deeper investigation and refined theories and hypotheses Critical Evaluation: Summary Remain aware of the research design (correlational vs. experimental) before assuming causality What was experimentally manipulated? If nothing was manipulated, what might be some other explanations for the relationship? # of Bars Look for more Population parsimonious hypotheses # of Churches About the Author These slides were extracted from a series of two PowerPoint presentations created by Owen Emlen [Email] Owen works as an independent consultant and is a member of BrainTech, LLC, where he creates online applications that analyze data that look for trends in behavior. Here is an example model