Types of research design experiments
- 1. Types of research design – experiments Chapter 8 in Babbie & Mouton (2001) Introduction to all research designs All research designs have specific objectives they strive for Have different strengths and limitations Have validity considerations
- 2. Validity considerations When we say that a knowledge claim (or proposition) is valid, we make a JUDGEMENT about the extent to which relevant evidence supports that claim to be true Is the interpretation of the evidence given the only possible one, or are there other plausible ones? "Plausible rival hypotheses" = potential alternative explanations/claims e.g. New York City's "zero tolerance" crime fighting strategy in the 1980s and 1990s - the reverse of the "broken windows" effect
- 3. The logic of causal social research in the controlled experiment Explanatory rather than descriptive Different from correlational research - one variable is manipulated (IV) and the effect of that manipulation observed on a second variable (DV) If … then …. E.g. "Animals respond aggressively to crowding" (causal) "People with premarital sexual experience have more stable marriages" (noncausal)
- 4. Three pairs of components: Independent and dependent variables Pre-testing and post-testing Experimental and control groups
- 5. Components Variables Dependent (DV) Independent (IV) Pre-testing and post-testing O X O Experimental and control groups To off-set the effects of the experiment itself; to detect effects of the experiment itself
- 6. The generic experimental design: R O 1 X O 2 R O 3 O 4 The IV is an active variable; it is manipulated The participants who receive one level of the IV are equivalent in all ways to those who receive other levels of the IV
- 7. Sampling 1. Selecting subjects to participate in the research Careful sampling to ensure that results can be generalized from sample to population The relationship found might only exist in the sample; need to ensure that it exists in the population Probability sampling techniques
- 8. Sampling 2. How the sample is divided into two or more groups is important to make the groups similar when they start off randomization - equal chance matching - similar to quota sampling procedures match the groups in terms of the most relevant variables; e.g. age, sex, and race
- 9. Variations on the standard experimental design One-shot case study X O No real comparison
- 10. A famous one-group posttest-only design Milgram's study on obedience Obedience to authority The willingness of subjects to follow E's orders to give painful electrical shocks to another subject A real, important issue here: how could "ordinary" citizens, like many Germans during the Nazi period, do these incredibly cruel and brutal things? If a person is under allegiance to a legitimate authority, under what conditions will the person defy the authority if s/he is asked to carry out actions clearly incompatible with basic moral standards?
- 11. One-group pre-test post-test design O 1 X O 2
- 12. Example We want to find out whether a family literacy programme enhances the cognitive development of preschool-age children. Find 20 families with a 4-year old child, enrol the family in a high-quality family literacy programme Administer a pretest to the 20 children - they score a mean of say 50 on the cognitive test The family participates in the programme for twelve months Administer a post-test to the 20 children; now they score 75 on the test - a gain of 25
- 13. Two claims/conclusions: 1 The children gained 25 points on average in terms of their cognitive performance 2 the family literacy programme caused the gain in scores VALIDITY - rival explanations
- 14. Static-group comparison X O O
- 15. Evaluating research (experiments) We know the structure of research We understand designs We know the requirements of "good" research Then we can evaluate a study Is it good? Can we believe its conclusions? Back to plausible rival hypotheses
- 16. Validity in designs If the design is not valid, then the conclusions drawn are not supported; it is like not doing research at all Validity of designs come in two parts: Internal validity can the design sustain the conclusions? External validity can the conclusions be generalized to the population?
- 17. Internal validity Each design is only capable of supporting certain types of conclusions e.g. only experiments can support conclusions about causality Says nothing about if the results can be applied to the real world (generalization) Generally, the more controlled the situation, the higher the internal validity The conclusions drawn from experimental results may not accurately reflect hat has gone on in the experiment itself
- 18. Sources of internal invalidity These sources often discussed as part of experiments, but can be applied to all designs (e.g. see reactivity) History Historical events may occur that will be confounded with the IV Especially in field research (compare the control in a laboratory, e.g. nonsense syllables in memory studies
- 19. Maturation Changes over time can be caused by a natural learning process People naturally grow older, tired, bored, over time
- 20. Testing (reactivity) People realize they are being studied, and respond the way they think is appropriate The very act of studying something may change it In qualitative research, the "on stage" effects
- 21. The Hawthorne studies Improved performance because of the researcher's presence - people became aware that they were in an experiment, or that they were given special treatment Especially for people who lack social contacts, e.g. residents of nursing homes, chronic mental patients
- 22. Placebo effect When a person expects a treatment or experience to change her/him, the person changes, even when the "treatment" is know to be inert or ineffective Medical research "The bedside manner", or the power of suggestion
- 23. Experimenter expectancy Pygmalion effect - self-fulfilling prophecies of e.g. teachers' expectancies about student achievement Experimenters may prejudge their results - experimenter bias Double blind experiments: Both the researcher and the research participant are "blind" to the purpose of the study. They don't know what treatment the participant is getting
- 24. Instrumentation Instruments with low reliability lead to inaccurate findings/missing phenomena e.g. human observers become more skilled over time (from pretest to posttest) and so report more accurate scores at later time points
- 25. Statistical regression to the mean Studying extreme scores can lead to inflated differences, which would not occur in moderate scorers
- 26. Selection biases Selection subjects for the study, and assigning them to E-group and C-group Look out for studies using volunteers
- 27. Attrition Sometimes called experimental (or subject) mortality If subjects drop out, it creates a bias to those who did not e.g. comparing the effectiveness of family therapy with discussion groups for treatment of drug addiction addicts with the worst prognosis more likely to drop out of the discussion group will make it look like family therapy does less well than discussion groups, because the "worst cases" were still in the family therapy group
- 28. Diffusion or imitation of treatments When subject can communicate to each other, pass on some information about the treatment (IV)
- 29. Compensation In real life, people may feel sorry for C-group who does not get "the treatment" - try to give them something extra e.g. compare usual day care for street children with an enhanced day treatment condition service providers may very well complain about inequity, and provide some enhanced service to the children receiving usual care
- 30. Compensatory rivalry C-group may "work harder" to compete better with the E-group
- 31. Demoralization Opposite to compensatory rivalry May feel deprived, and give up e.g. giving unemployed high school dropouts a second chance at completing matric via a special education programme if we assign some of them to a control group, who receive "no treatment", they may very well become profoundly demoralized
- 32. External validity Can the findings of the study be generalized? Do they speak only of our sample, or of a wider group? To what populations, settings, treatment variables (IV's), and measurement variables can the finding be generalized?
- 33. External validity Mainly questions about three aspects: Research participants Independent variables, or manipulations Dependent variables, or outcomes Says nothing about the truth of the result that we are generalizing External validity only has meaning once the internal validity of a study has been established Internal validity is the basic minimum without which an experiment is uninterpretable
- 34. External validity Our interest in answering research questions is rarely restricted to the specific situation studied - our interest is in the variables, not the specific details of a piece of research But studies differ in many ways, even if they study the same variables: operational definitions of the variables subject population studied procedural details observers settings Generally bigger samples with valid measures lead to better external validity
- 35. Sources of external invalidity Subject selection - Selecting a sample which does not represent the population well, will prevent generalization Interaction between the testing situation and the experimental stimulus When people have been sensitized to the issues by the pre-test Respond differently to the questionnaires the second time (post-test) Operationalization
- 36. Operationalization We take a variable with wide scope and operationalize it in a narrow fashion Will we find the same results with a different operationalization of the same variable?
- 37. Field experiments "natural" - e.g. disaster research Static-group comparison type Non-equivalent experimental and control groups
- 38. Strengths and weaknesses Strengths Control Manipulating the IV Sorting out extraneous variables Weaknesses Articifiality - a generalization problem Expense Limited range of questions
- 39. IN CONCLUSION Donald Campbell often cited Neurath's metaphor: "in science we are like sailors who must repair a rotting ship while it is afloat at sea. We depend on the relative soundness of all other planks while we replace a particularly weak one. Each of the planks we now depend on we will in turn have to replace. No one of them is a foundation, nor point of certainty, no one of them is incorrigible"