Lecture
An essential element of planning is the way in which the groups are formed. The most rigorous criterion requires randomization, a random order of forming the groups.
Inter-individual differences, or differences between subjects and inequality between groups, are the main source of threats to internal validity in between-group schemes. They may affect the dependent variable or interact with the independent variable, which distorts the form of the causal dependence under study.
Sometimes increasing the samples is considered as a way of controlling non-systematic confoundings. But the problem of the non-equivalence of samples is not thereby solved, because:
- even with randomization, people who are «similar» in some property may end up in different groups;
- with large n the law of large numbers comes into play – the more participants there are in the experiment, the greater the probability that the extraneous variables will affect the difference (the more confoundings there are).
Types of between-group designs (comparison schemes):
RXO’
R O’’
Randomization – R – here acts as a strategy for selecting participants into groups from the population; it allows the groups to be regarded as equivalent and the difference in the dependent variable to be attributed specifically to the action of the independent variable. The experimental effect for between-group schemes is the result of comparing the sample values of the dependent variable in the experimental and control groups.
A more complex variant is the randomized block design. It is used when subgroups are formed within each group.
The experimental and control groups can be equated with respect to confounding with block extraneous variables («knowledge about the experiment», pupils of different classes, etc.) by assigning participants from each block to the experimental and control groups in random order.
Pretesting partially solves the problem of accounting for and controlling the initial level of the dependent variable, but the non-equivalence of the groups may be hidden in unmeasured variables.
2. A design with pretesting and posttesting and a control group.
RO’XO”
RO O”
The main effect here is computed as a comparison of the difference in the increments of the dependent variable between the experimental and control conditions (O2-O1)- (O4-O3). Such designs may have insufficient external validity.
3. The Solomon design.
A comparison of four groups: two experimental and two control, with the introduction of the factor «presence or absence of pretesting».
RO1XO2
RO3 O4
R XO5
R O6
The factor of inter-individual differences is controlled by strategies for selecting and matching participants. The aspect of the representativeness of participants is external validity; the equivalence of groups is internal validity.
Matching takes place when participants are not selected from the population but are distributed into groups from an already existing pool of participants.
That is, here the condition of randomized selection is not met. In the case of the randomization strategy for selection from a population (see above), the external and internal validity of the experiment turn out to be interrelated.
4. The strategy of stratified selection or matching of participants (the strategy of random distribution of strata).
1) Forming groups corresponding to the identified characteristics of the strata (sex, age, etc.). There will be as many such groups as there are levels identified according to the changes in the extraneous variable.
2) Individuals from each stratum are randomly selected into the experimental and control groups. There will be as many experimental groups as there are levels of the independent variable. As a result, each level of the extraneous variable is equally represented in each of the selected (or matched) groups.
This strategy is often used to control an additional variable.
5. The strategy of pairwise matching.
Used in the case of matching, when the sample of participants has already been determined and they can be subjected to pretesting.
1) Determine the pair of participants with the most pronounced values of the extraneous variable (and then those with less pronounced values, then with even less pronounced ones…)
2) Distribute the participants of each pair between the two groups according to a random strategy or an even-odd strategy.
This strategy is mainly used with small samples.
6. Random selection of groups.
Used when approximating an experiment of full correspondence presupposes implementing the X-influences under the real conditions of life activity. For example, comparing different methods of instruction.
Comments