Chapter 1: Variables and Research Design

1.2 Variables

Variable: something that can vary; it can take on many different values or categories

Examples: gender, typing speed, top speed of a car, number of reported symptoms of an illness, temperature, attendances at rock festivals, level of anxiety

1.2.1 Characteristics of Variables

Continuous: variables that can take ^^any value within a given range^^; the variable itself doesn’t change in discrete jumps

• Example: temperature
• there is an assumption that the underlying variable itself is continuous, even if the way in which we measure it is not
  • Example: level of anxiety

Discrete: variables can take on ^^only certain discrete values within the range^^

• Example: reported number of symptoms of an illness that a person has

Categorical: values that the variables can take are ^^categories^^

• Example: gender

• can sometimes have many possible values

  • Example: type of occupation (e.g., judge, teacher, miner, etc.)

1.2.2 Dichotomizing Continuous and Discrete Variables

• Dichotomizing (dividing into 2 categories) of continuous and discrete variables is quite common in psychology as it enables us to find out if there are differences between groups who may be at the extremes of the continuous or discrete variables.
• Such practice is ^^not recommended^^ as it ^^reduces the sensitivity of your statistical analysis^^.
• Streiner (2002) suggests that it would be better to think of ^^variables as being on a continuum^^. We should measure such constructs on continua rather than dichotomizing them.
• The practice of dichotomizing continuous variables tends to lead to ^^research that is low in power.^^ The reason for this is that it results in us ^^losing a lot of information about participants^^.
• Streiner highlights research that has shown that analyses using dichotomous variables are about ^^67% as efficient^^ as analyses using the original continuous/discrete measures.
• Maxwell and Delaney (1993) have shown that dichotomizing continuous variables can actually lead to ^^spurious findings arising from statistical analyses^^.

Levels of Measurement

Nominal Scales: consist of ^^categories that are not ordered^^ in any particular way

• Example: gender (male or female)
• simply assigning people to categories and the data we obtain are in the form of ^^frequency counts^^ (i.e., tells us how many people we have in each category)

Ordinal Scales: have some sort of ^^order in the categories^^ (e.g., in terms of magnitude) but the intervals between adjacent points on the scale are not necessarily equal

Interval Scales: have ^^equal intervals^^ between adjacent scores but ^^do not have an absolute zero^^

• Example: temperature (zero point does not equate to zero temperature)

Ratio Scales: have ^^equal intervals^^ between adjacent scores on the scale and an absolute zero

• Example: speed (zero point means zero speed)

Research Designs

Extraneous and Confounding Variables

Extraneous Variables: ^^variables that might have an impact on the other variables^^ that we are interested in but we may have ^^failed to take these into account^^ when designing our study

• If extraneous variables are overlooked, the conclusions that may be drawn from the studies may be unreliable

Confounding Variable: a specific type of extraneous variable that is related to both of the main variables that we are interested in

Correlational Designs

Correlational Designs: those that ^^investigate relationships between variables^^

• It is difficult to establish causal relationships in correlational designs because we are simply ^^observing and recording changes in variables^^ and trying to establish whether they co-vary in some meaningful way

We cannot infer causation from correlations.

Causation

• To establish a causal relationship, we need to ^^manipulate one variable^^ (change it systematically) and then see what effect this has on the other variables

The Experimental Design

Experimental Design: those where the experimenter ^^manipulates one variable (IV) to see what effect this has upon another variable (DV)^^; we are usually looking for ^^differences between conditions of the IV^^; a hallmark of this design is ^^random allocation^^ of participants to the conditions of the IV

Independent Variable (IV): the variable ^^manipulated by the experimenter^^; its value is not dependent upon (is independent of) the other variables being investigated

Dependent Variable (DV): it is assumed to be ^^dependent upon the value of the IV^^

Research Hypothesis: our ^^prediction^^ of how specific variables might be related to one another or how groups of participants might be different from each other

Random Allocation: one of the major defining features of an experimental design; if we randomly allocate participants to conditions, we can be more confident in our ability to infer a causal relationship between the IV and the DV

Quasi-Experimental Designs

Quasi-Experimental Design: involve seeing if there are differences on the DV between conditions of the IV; there is ^^not random allocation^^ of participants to the various conditions of the IV

• One of the problems of this design is that, because participants are not randomly allocated to the various conditions that make up the IV, we cannot be certain that our manipulation of the IV (or pseudo-manipulation) is responsible for any differences between the various conditions

Between-Participants and Within-Participants Designs

Within-Participants Designs

Within-Participants Designs: have the ^^same participants in every condition^^ of the IV; each participant performs under all conditions of the study

• Advantages:
  • you are able to control for many inter-individual confounding variables
  • you need to find fewer participants in order to complete your research
  • this design tend to have more statistical power than between-participant designs; they are more likely to detect an effect that we are looking for in the population
• Disadvantages:
  • Order Effects: a consequence of within-participants designs whereby ^^completing the conditions in a particular order leads to differences in the DV^^ that are not a result of the manipulation of the IV; differences between the conditions of the IV ^^might be due to practice, fatigue, or boredom^^ rather than to the experimenter’s manipulation of the IV
  • Ways to eliminate order effects:
    • introduce counterbalancing into your design (i.e., get 1/2 of the participants to complete the first condition followed by the second condition and the other half to complete the second condition first followed by the first condition)
    • Counterbalancing: where you ^^systematically vary the order in which participants take part in the various conditions of the IV^^; this would be introduced into a study where you have a ^^within-participants design^^
  • having participants take part in both conditions means that they are more likely to realize the purpose of the experiment
  • Demand Effects: participants want to do what the experimenter wants them to and so may perform how they believe they should do rather than how they would normally have done
  • you ^^cannot use them in many quasi-experimental designs^^

Between-Participants Designs

Between-Participants Designs: have ^^different groups of participants in each condition of the IV^^; thus, the group of participants in one condition of the IV is different from the participants in another condition of the IV

• Advantages:
  • Because you have different groups of participants in each condition of the IV, each participant is less likely to get bored, tired, or frustrated with the study.
  • Your research is going to be less susceptible to practice effects and the participants are less likely to work out the rationale of the study.
  • This reduces order and demand effects, and you can, to a large extent, eliminated these factors as extraneous variables from your study
• Disadvantages:
  • You will need more participants than you would for a completely within-participants design.
  • If you use different participants in each condition, you lose a certain degree of control over the inter-participant confounding variables.