Psy 232 GF08 Experimental Designs:  Between Groups

Scores from separate groups of individuals are compared

           Step 1:  A sample of participants is selected from the population

                       NOTE regarding N:  sample should be sufficiently large, so that when

                         groups are formed you’ll have 30+ per group if possible

                        The CLT (ensures normality assumption) applies to your groups,

                         not to the overall sample!

           Step 2:  Participants assigned to different groups (i.e., to different “treatments”)

           Step 3:  Scores from the treatment groups are collected.

           Step 4:  Scores are analyzed via appropriate statistics tests.

           Note:  A between-groups design allows only one score per person.

                     Thus # scores = # participants.

Advantages of independent-measures designs

NOTE: ‘independence’ à random assignment & other methodological steps to ensure it.

          Scores are independent of one another, since each participant is measured

          only once in only one treatment.

                That is, carry-over effects from one tx to another one are eliminated.

                Fatigue effects from participating in a series of treatments are eliminated.

                Contrast effects between treatments are eliminated

          These designs can be used in almost all research situations

               (although they aren’t always the best ones)

Disadvantages of independent-measures designs

          Primary problem – large individual differences              

                     Can become confounding variables [Examples . . .]

                     Can produce high variability in the scores, thus obscuring effects of  treatments

                     In t-test, this would increase s_M, thus causing “chance” to be larger.

          Other problem – large # of participants are needed

                    Increases testing-time and related hassles for the researcher

                    May be difficult to find participants when studying special populations

Eliminating confounding variables

          Process for selecting participants should be similar for all groups

          All aspects of participants’ experiences should be the same,

                   except for experimental manipulation.

          Participants should be similar across all groups.

          Primary technique:  random assignment of participants to groups.

                     Note:  this can be restricted as needed to ensure equal #’s per group.

                     Note:  this works particularly well with large n’s, but less well with small n’s

          Additional techniques:

                      Matching participants on particular vars to ensure similarity across groups

                       This requires prior measurement of the variable of interest

                        May be costly or tedious process

                        May be difficult to accomplish when more than one variable is of concern

                        Holding a variable constant, or else restricting its range of variation

                            Advantage – ensures that participants are similar on the var of concern.

                            Disadvantage –limits external validity since generalization is limited.

          Most of the time:  when possible, go with larger n’s and use random assignment.

Managing effects of high variability

     Remember that stats tests typically compare treatment-effects to chance-effects

          If the treatment-effects = chance-effects, then we say that the tx did not work.

          If the treatment-effects ≠ chance-effects, then we say that the treatment did work.

          Thus we usually want to increase treatment-effects and to minimize chance-effects.

          Chance-effects are due to differences within groups (i.e., s_M).

          Since individual diffs are part of “chance,” we reduce them as much as possible.

                Techniques for reducing the effects of individual differences:

                       Standardize research materials procedures.  [Examples . . .]

                       Standardize research settings.  [Examples . . .]

                       Hold a participant variable constant.  [Examples . . .]

                       Increase sample size.  [Examples . . .]

                           Note:  this doesn’t reduce individual diffs, but it does reduce their effect.                                        

                                       See t-test formula for reminder of why this is true

Typical designs

Two-group design – One IV with two levels:  often a “treatment” versus a “control”

          Advantage 1:  Simple to set up and carry out

          Advantage 2:  Useful for comparing extreme values of an IV (e.g. zero versus max).

          Disadvantage 1:  Provides info about only two pts, so rel betw DV & IV is unclear.

                   [Examples:  when the relation for all values is curvilinear . . .]

          Disadvantage 2:  More than one control is often needed . . .

                   [Examples: “no-treatment” versus “placebo” versus treatment . . .]

Multi-group design – One IV with three or more levels (treatments)

           Multiple control groups are possible.  [Examples . . .]

           Multiple treatment groups are possible.  [Examples . . .]

           Gives a clearer picture of the overall rel between the DV and IV [Examples . . .]

           Minor caution:  using too many groups may cause some analysis problems.