CHAPTER 4 Causal Designs and Marketing Experiments © 2013 Cengage Learning. All Rights Reserved. May not be scanned, copied or duplicated, or posted to a publicly accessible website, in whole or in part. The Quest for Causality “Are there other possible factors that could have led to the observed changes?” scientific concept of causality (“probabilistic causation”): a number of determining conditions act together to make an effect probable causality can only be inferred, never proven 3 conditions for causal inferences: concomitant variation – correlation that helps rule out: • reverse causation • omitted variables • insufficient variation 2. time order of occurrence of variables – precedence 3. elimination of other possible causal factors 1. © 2013 Cengage Learning. All Rights Reserved. May not be scanned, copied or duplicated, or posted to a publicly accessible website, in whole or in part. Experimentation experiment: the manipulation of one or more independent variable in a planned fashion and measurement of the effects on the dependent variable the fundamental research tool used to help identify causal relationships objective is to measure the effect of explanatory (independent) variables on a variable of interest, while controlling for other variables © 2013 Cengage Learning. All Rights Reserved. May not be scanned, copied or duplicated, or posted to a publicly accessible website, in whole or in part. Basic Definitions in Experimentation treatments – the independent variables that are manipulated and whose effects are then measured test units – recipients of the treatments, whose response is measured dependent variables – measures taken on the test units extraneous variables – all variables, other than the treatments, that potentially affect the dependent variable experimental design – specification of treatments, test units, dependent variables, and procedures for dealing with extraneous variables © 2013 Cengage Learning. All Rights Reserved. May not be scanned, copied or duplicated, or posted to a publicly accessible website, in whole or in part. Basic Definitions in Experimentation (continued) validity in experimentation internal validity – minimum validity to make conclusion: have effects of extraneous variables been accounted for? external validity – can the results be generalized? X-O-R syntax – standard convention of symbols used to define experiments: X – exposure of test group to treatment O – observation or measurement of the dependent variable R – test units have been assigned at random to treatments movement from left to right indicates passage through time symbols in a horizontal row refer to a specific treatment group symbols aligned vertically refer to the same moment in time © 2013 Cengage Learning. All Rights Reserved. May not be scanned, copied or duplicated, or posted to a publicly accessible website, in whole or in part. Types of Extraneous Variables history effect – events concurrent with the experiment that may affect the dependent variable maturation effect – effect of changes in the test units over time testing effect – effect of taking a measure before treatment main testing effect – effect of 1st observation on the 2nd observation interactive testing effect – effect of pre-treatment observation on the response to the treatment © 2013 Cengage Learning. All Rights Reserved. May not be scanned, copied or duplicated, or posted to a publicly accessible website, in whole or in part. Types of Extraneous Variables (continued) instrumentation – changes in the observers or calibration of the measuring instrument statistical regression effect – effect of test units being selected for treatment on the basis of an extreme pretreatment score selection bias – effect of treatment groups differing on the dependent variable prior to the presentation of the treatments test unit mortality – effect of test units withdrawing from the experiment while it is in progress © 2013 Cengage Learning. All Rights Reserved. May not be scanned, copied or duplicated, or posted to a publicly accessible website, in whole or in part. Three Pre-Experimental Designs One-Shot Case Study (X O) – test units exposed to treatment X, then measurement taken One-Group Pretest-Posttest Design (O1 X O2) – pretest measurement is added to the one-shot case study design Static-Group Comparison – two treatment groups, one exposed to treatment and control group that is not, both observed only after treatment Experimental group: Control group: X1 X2 O1 O2 © 2013 Cengage Learning. All Rights Reserved. May not be scanned, copied or duplicated, or posted to a publicly accessible website, in whole or in part. Three True Experimental Designs Pre-test Post-test Control Group Design – controls for extraneous variables, but not interactive testing effect experimental group: R control group: R (O2 – O1) – (O4 – O3) = TE + IT O1 O3 X1 O2 O4 Solomon Four-Group Design – controls for extraneous variables and interactive testing effect with 2nd control group Experimental group 1: R O1 Control group 1: R O3 Experimental group 2: R X O2 O4 X Control group 2: R TE = O5 - O6, but all results are used equally © 2013 Cengage Learning. All Rights Reserved. May not be scanned, copied or duplicated, or posted to a publicly accessible website, in whole or in part. O5 O6 Three True Experimental Designs (continued) Post-test-Only Control Group Design – last two groups of the Solomon four-group design since there is no pre-test, the interactive testing effect cannot occur Experimental group: R Control group: R X O1 O2 O1 - O2 = TE true experimental designs control for extraneous variables (EXT) © 2013 Cengage Learning. All Rights Reserved. May not be scanned, copied or duplicated, or posted to a publicly accessible website, in whole or in part. Quasi-Experimentation control over data collection procedures no randomization of assignment of treatments little control over scheduling of the treatments designed to increase external validity much more susceptible to confounded results time-series experiment periodic measurement before and after treatment O1 O2 O3 O4 X O5 O6 O7 O8 weaknesses • inability to control history effect • possibility of an interactive testing effect from repeated measurements © 2013 Cengage Learning. All Rights Reserved. May not be scanned, copied or duplicated, or posted to a publicly accessible website, in whole or in part. Quasi-Experimentation (continued) multiple time-series design a time-series design with control group of test units experimental group: control group: O O O O O O X O O O O O O problems: • difficulty of selecting a well-matched control group • possible interactive effect in the experimental group equivalent time-sample design experimental group acts as its own control, preferably when the effect of the treatment is transient: O X1 O X0 OOO X1 OO X0 O (X1 = treatment and X0 = absence of treatment) possible interactive testing effect © 2013 Cengage Learning. All Rights Reserved. May not be scanned, copied or duplicated, or posted to a publicly accessible website, in whole or in part. Quasi-Experimentation (continued) nonequivalent control group design both experimental and control groups are measured pre-test and post-test, but the control group is not equivalent to the test group experimental group: O1 X O2 control group: O3 O4 control over who is exposed to the treatment, thus improving the ability to control the effects of history, maturation, main testing, instrumentation, selection, and test unit mortality possible interactive testing effect possible regression effect if either group selected on the basis of extreme scores © 2013 Cengage Learning. All Rights Reserved. May not be scanned, copied or duplicated, or posted to a publicly accessible website, in whole or in part. Managerial Aspects of Experimentation and Quasi Experimentation descriptive techniques (secondary data, observation, surveys, panels, web log analysis and simulation) measure correlation, not causation laboratory versus field environments laboratory provides better control of confounds, greater internal validity field experiments provide greater external validity laboratory generally less expensive laboratory often requires less time © 2013 Cengage Learning. All Rights Reserved. May not be scanned, copied or duplicated, or posted to a publicly accessible website, in whole or in part. Control of Invalidity + = factor controlled ? = possible concern History Maturation Testing Instrumentation Regression Selection Mortality Interactive testing effect – – – – – – ? – – static group comparison – = weakness one group pretest posttest one shot Pre-experimental designs + ? + + + – – – © 2013 Cengage Learning. All Rights Reserved. May not be scanned, copied or duplicated, or posted to a publicly accessible website, in whole or in part. Control of Invalidity (continued) – = weakness + = factor controlled ? = possible concern History Maturation Testing Instrumentation Regression Selection Mortality Interactive testing effect + + + + + + + – Posttest-only control group Pretestposttest control group Solomon fourgroup True experimental designs + + + + + + + + © 2013 Cengage Learning. All Rights Reserved. May not be scanned, copied or duplicated, or posted to a publicly accessible website, in whole or in part. + + + + + + + + Control of Invalidity (continued) History Maturation Testing Instrumentation Regression Selection Mortality Interactive testing effect Nonequivalent control group ? = possible concern Equivalent time sample + = factor controlled Multiple time series – = weakness Time series Quasi-experimental designs – + + ? + + + – + + + + + + + – + + + + + + + – + + + + ? + + – © 2013 Cengage Learning. All Rights Reserved. May not be scanned, copied or duplicated, or posted to a publicly accessible website, in whole or in part. Limitations of Experimentation can not always control effects of extraneous variables, particularly in field experiments lack of cooperation in field can limit experimental activity lack of knowledge about experimental procedures on the part of marketing personnel can be costly, time-consuming and requires expert data analysis responses can still be biased by experimenter © 2013 Cengage Learning. All Rights Reserved. May not be scanned, copied or duplicated, or posted to a publicly accessible website, in whole or in part. Stages in Conducting an Experiment 1. State problem 2. Formulate hypothesis 3. Construct experimental design 4. Mock up data – will it meet information needs? 5. Check results can be analyzed by available statistical procedures 6. Perform experiment 7. Apply statistical analysis procedures 8. Draw conclusions © 2013 Cengage Learning. All Rights Reserved. May not be scanned, copied or duplicated, or posted to a publicly accessible website, in whole or in part. Four Design Procedures: an Overview determine statistical significance – is “signal” (measured effect) greater than “noise” (sampling error)? completely randomized design (CRD) one nominally scaled independent variable treatments assigned randomly randomized block design (RBD) – combines test units into blocks based on extraneous variable, reducing sampling error each treatment must appear at least once in each block number of test units number of treatments © 2013 Cengage Learning. All Rights Reserved. May not be scanned, copied or duplicated, or posted to a publicly accessible website, in whole or in part. Four Design Procedures: an Overview (continued) Latin square (LS) design – two extraneous variables number of categories of each variable to be controlled exactly equals the number of treatments treatments assigned to cells randomly factorial design – for more than one independent variable or interactive effects* between variables levels = independent variable categories treatments = combinations of levels allows separation of interaction and main effect through statistical procedures * relationship between independent and dependent variable is different for different categories of another independent variable © 2013 Cengage Learning. All Rights Reserved. May not be scanned, copied or duplicated, or posted to a publicly accessible website, in whole or in part. International Marketing Experiments Hold equally well for domestic and for international research: principles of causality structure of experimental design nature of marketplace quasi-experiments © 2013 Cengage Learning. All Rights Reserved. May not be scanned, copied or duplicated, or posted to a publicly accessible website, in whole or in part.