Basic Experimental Design • Common Problems • Assigning Participants to Groups • Single variable experiments – bivalent – multivalent – baseline • Multivariate – factorial – converging series © 2001 Dr. Laura Snodgrass, Ph.D. 1 Common Problems • • • • Confounds Lack of control group(s) Nonequivalent control groups Why control groups – history – maturation – testing – instrument decay – statistical regression © 2001 Dr. Laura Snodgrass, Ph.D. 2 Assigning Participants to Groups • Independent or Random Groups Design – between groups • Repeated Measures – within groups © 2001 Dr. Laura Snodgrass, Ph.D. 3 Between Groups • Advantages – generalizable – collect more data at a given level – shorter time for each participant • Disadvantages – may not be random – unequal N – potential confounds – requires more participants © 2001 Dr. Laura Snodgrass, Ph.D. 4 Between Groups • Matching to equate groups and decrease error variance • How – correlated variables – pairs – yoked controls – performance criterion © 2001 Dr. Laura Snodgrass, Ph.D. 5 Matching • Advantages – equates groups – increase power of experiment – decrease number of participants needed • Disadvantages – extra work – extra testing – lose individual differences - less generalizable © 2001 Dr. Laura Snodgrass, Ph.D. 6 Repeated Measures • Advantages – fewer participants needed – impt for special groups – statistically more powerful • Disadvantages – not naïve after first trials – order effects • practice and fatigue • non-symmetric or differential transfer © 2001 Dr. Laura Snodgrass, Ph.D. 7 Counterbalancing • Vary order of treatment to distribute or measure order effects • Complete counterbalancing – within participants ABBA – between AB for some, BA for others • Latin Squares – each cond at each ordinal position – precedes and follows each other once © 2001 Dr. Laura Snodgrass, Ph.D. 8 Counterbalancing • Randomized blocks • Time interval between trials – mortality © 2001 Dr. Laura Snodgrass, Ph.D. 9 Single Variable Experiments • Bivalent – one independent variable with two levels • Multivalent (functional) – one independent variable with three or more levels • Baseline © 2001 Dr. Laura Snodgrass, Ph.D. 10 Bivalent • Two levels of the independent variable – experimental and control groups – two different levels of the variable • Post-test only vs. pre-test/post-test • Advantages – easy to interpret and analyze – decide if IV is worth studying © 2001 Dr. Laura Snodgrass, Ph.D. 11 Bivalent • Disadvantages – limited theoretical value – conclusions may be based on arbitrary choice of levels – negative findings are not conclusive – does not describe shape of relationship therefore you may over generalize for non-linear relationships • interpolation and extrapolation • plateau or asymptote © 2001 Dr. Laura Snodgrass, Ph.D. 12 Multivalent (functional) • Gives more info about the shape of the relationship • Advantages – better estimate true relationship – individual choice of levels becomes less critical • Disadvantages – more: time, effort, cost, subjects – more complex statistics and interpretation © 2001 Dr. Laura Snodgrass, Ph.D. 13 Baseline • Only works with certain types of variables – will not work with variables that cause permanent change • Procedure: – establish baseline or steady-state response level – introduce IV until stable transition – allow subject to return to baseline © 2001 Dr. Laura Snodgrass, Ph.D. 14 Baseline • Advantages – rules out most confounds – easy to interpret (often no statistics) – flexible and replicable – investigate behavior of an individual • Disadvantages – does not show small changes – may not generalize © 2001 Dr. Laura Snodgrass, Ph.D. 15 Multivariate Experiments • Factorial Designs – two or more independent variables, each with two or more levels – variables can be all between, all within, or mixed in many combinations • Converging series – series of small experiments in which a variable manipulated in an earlier experiment becomes a control variable in a later experiment © 2001 Dr. Laura Snodgrass, Ph.D. 16 Factorial • Design matrix – produces a family of functions – study main effects and interactions • Advantages – study interactions – increases precision and generalizability – decrease statistical error and increase power – theoretical value © 2001 Dr. Laura Snodgrass, Ph.D. 17 Factorial • Disadvantages – increases time, money and number of subjects increases dramatically as number of cells increases – assumptions of ANOVA may not be met – N-way interactions are very difficult to interpret © 2001 Dr. Laura Snodgrass, Ph.D. 18 Converging series for applied problems • Optimal designs – e.g. car, medical treatment, office • Find an optimal level of a variable and turn it into a control variable – lose higher order interactions © 2001 Dr. Laura Snodgrass, Ph.D. 19 Converging Operations • Converge on a single hypothesis – start with several possible hypotheses or explanations – each experiment eliminates one or more until only one remains (hopefully) • For example: – perceptual defense against vulgar words – isolation tank © 2001 Dr. Laura Snodgrass, Ph.D. 20 Converging Series • Advantages – flexible, many choice points – efficient, leave out factors that have no effect – built in replications • Disadvantages – interactions are lost – almost always between subjects – analyze and interpret prior before next experiment so can take a long time © 2001 Dr. Laura Snodgrass, Ph.D. 21