1 Multiple-choice example. 2 Solution Since the IV is manipulated by the experimenter, its not independent of any other variable. A is false. C is also.

1

Multiple-choice example

2

Solution

• Since the IV is manipulated by the experimenter, it’s not independent of any other variable. A is false.

• C is also false, because the IV is manipulated by the experimenter. • D is false, because the IV is supposed to influence the DV, not vice

versa. • B is correct: the experimenter manipulates the IV.

3

4

Solution

• A variable is a SET of conditions – one condition cannot be a variable. A is false.

• Recognition time is supposed to be affected by the visual field in which words are presented. Recognition time is the dependent variable. B is false.

• The visual field cannot be the dependent variable, because it is manipulated directly by the experimenter. D is false.

• The correct answer is C.

5

Lecture 2

PSYCHOLOGICAL EXPERIMENTS

6

Last week …

• I reviewed three research strategies:

1.Experimental

2.Correlational

3.Obervational

• The primary aim of the first two types is to show CAUSAL RELATIONSHIPS among variables.

7

A causal model

We run an EXPERIMENT to establish that the INDEPENDENT VARIABLE (IV) has a causal effect upon the DEPENDENT VARIABLE (DV).

8

Experimental control

By CONTROLLING the independent variable (IV) and negating the effects of EXTRANEOUS VARIABLES upon the dependent variable (DV), the experimenter hopes to show that the IV has a CAUSAL EFFECT upon the DV.

9

Comparison

• In a true experiment, there are at least TWO conditions.

1.an active or EXPERIMENTAL condition;

2.a comparison or CONTROL condition.

• The two conditions must be identical, apart from the absence of the active ingredient in the control condition.

• This is the RULE OF ONE VARIABLE.

10

Individual differences

• Suppose you want to show that training improves performance of a skilled task.

• You compare trained and untrained participants in their performance.

• But people vary enormously in their skill anyway.

• You may find that the best performer is in the untrained group; and the worst may be in the trained group!

11

Random assignment

• One solution to the problem of individual differences is to assign participants AT RANDOM to the experimental and control conditions.

• John may be very skilful, Joe may be very unskilful and Fred may be average.

• Assigning participants at random, however, should result in an EQUAL MIX of Johns, Freds and Joes in the experimental and control groups.

12

Random assignment…

• The purpose of random assignment is to avoid BIAS – a tendency for Freds, Joes or Mary’s to be in one group, rather than the other.

• If there is such a bias, the groups will not be COMPARABLE and individual differences will be an EXTRANEOUS or CONFOUNDING VARIABLE in the study.

13

An ‘event’

• An EVENT is the outcome of an experiment of chance, such as rolling a die, tossing a coin – or running a psychological experiment.

• Chance is an important factor in the outcome of an experiment.

• Joe, Fred and Mary participated this time; but Anne, Jim and Fiona could easily have done so – and their scores would almost certainly have been different.

14

Probability

• The PROBABILITY of an event is a measure of its likelihood, which can take values from zero (an impossible event) to unity (a certainty).

• There have been several definitions of probability.

• Each of them is, from a philosophical point of view, deeply problematic.

15

Classical definition

• The probability of an event is the NUMBER OF WAYS in which the event can occur, divided by the TOTAL NUMBER OF OUTCOMES.

• Roll a die. • What is the probability of a six? • There is ONE way of getting a six. There

are SIX possible outcomes. • So the probability of a six is 1/6.

16

Examples

• Roll a die. What is the probability of an even number?• That could happen in three ways: 2 spots, 4 spots or six

spots. • So the probability is 3/6 = ½. • What is the probability of a seven? There is NO WAY in

which that could happen, so the probability is 0/6 = 0 (indicating an IMPOSSIBILITY).

• A number between 1 and 6? That event could happen in six ways, so the probability is 6/6 = 1 (indicating a CERTAINTY).

17

The classical definition

18

‘random’

• The everyday meaning of this word is ‘unpredictable’.

• In statistics, however, when they say that someone has been chosen ‘at random’, they mean that the action can be described in terms of PROBABILITY.

• ‘A person was selected at random from a pool of 60 volunteers.’

• This means that every volunteer had a one in sixty chance of being selected: i.e., that person’s probability of selection was 1/60.

19

Random assignment

• Random assignment isn’t as easy as it seems.

• Suppose you have fifty participants whom you want to assign at random to two groups, Group A and Group B.

• How would you do it?

20

Between subjects experiments

• ‘Subjects’ was the old word for the modern PC term ‘participants’.

• The old term persists in the taxonomy of experimental designs.

• Participants are assigned AT RANDOM to the experimental and control conditions.

21

Large group experimental strategy

• The rationale of random assignment to groups is that individual differences will ‘cancel out’.

• For that to happen, the groups must be sufficiently LARGE.

• There are good theoretical reasons for this, which I shall touch upon later.

22

Within subjects experiments

• We can test each person under ALL CONDITIONS.

• This is known as the WITHIN SUBJECTS or REPEATED MEASURES strategy.

• If we do that, each participant serves as his or her OWN CONTROL.

• This is an alternative solution to the problem of individual differences.

23

A within subjects experiment

• Stroop (1935) tested the SAME GROUP OF PARTICIPANTS under all conditions.

• So each person was serving as his or her own control.

24

The results

• Participants took longer to name colours than to read colour words printed in black and white.

• But they took MUCH longer to name the print colours of the colour words in the ‘conflicting’ condition.

25

Advantages of within subjects experiments

• The within subjects experiment has several advantages over the between subjects experiment.

• You cannot argue that people of different abilities were tested under different conditions.

• You produce as much data with fewer participants.

• You make maximum use of the participants while they are available.

26

Order effects

• On the other hand, the performance of a task may be affected by the order in which the tasks are performed.

• Perhaps it’s easier to name the colours of non-word objects after you have tried the conflicting colour-word task than after you have tried the reading task?

• If all participants were to perform the three tasks in the same order, therefore, the results might be confounded with ORDER EFFECTS.

27

Counterbalancing

• We vary the order of presentation, so that each task is presented equally often in First, Second and Third positions.

• This procedure is known as COUNTERBALANCING.

28

Cyclic permutation

• Counterbalancing can be achieved by CYCLIC PERMUTATION.

• A, B, C and D represent the different conditions in the experiment.

• The condition on the right is continually moved to the leftmost position each time, so that all the other conditions slide ‘to the right’.

• Each condition occurs in the same serial position with approximately equal frequency.

29

Single-subject experiments

• There’s an alternative strategy to the LARGE-GROUP approach.

• It is quite possible to run an experiment on ONE PARTICIPANT.

• A therapist observes the frequency of a child’s self-harming actions over a period of time, counting the number of actions during each of 30 sample periods.

• Over trials 10 to 20, the therapist intervenes with a new technique.

30

Evidence for causality

• First, we establish the BASELINE performance level.

• Over the intervention period, there is a marked drop in the incidence of self-harming.

• When the intervention ceases, the self-harming rises to the pre-intervention baseline level.

• We can conclude that the intervention has caused the reduction in self-harming.

31

Issues with single-subject experiments

• Single-subjects experiments and those of similar design with very few subjects (‘small-n’ research) are frequently used in clinical research.

• They afford strong evidence of causality IN THE CASES CONCERNED.

• Can we generalise to ALL possible cases? • In our present example, can we conclude that

the intervention will be useful with ALL self-harmers?

32

A study of smoking

• Does smoking shorten life? • A researcher gathers information about regular

smokers and non-smokers, and compares their average age at death.

• There are many problems with this kind of study. • They all stem from the fact that the people

studied were NOT RANDOMLY ASSIGNED to the smoking and non-smoking conditions: these participants were ‘self-selecting’.

33

Control by sampling

• There are differences in socioeconomic status between smokers and non-smokers.

• There are occupational and lifestyle differences.

• There are differences in the proportions of men and women who smoke.

• All the researcher can do is to try to control the potential confounding variables by SAMPLING appropriately.

34

Control by sampling …

• The researcher will make sure that the GENDER mix is equal in the smokers and non-smokers.

• The INCOMES of the two groups must be comparable.

• The EDUCATIONAL LEVELS of the smokers and the non-smokers must be the same.

• These variables are all associated (CORRELATED) with age at death.

• But so are many other variables.

35

Quasi-experiments

• The smoking study is an attempt to approximate a true experiment by trying to ensure that the smoking and non-smoking groups are as COMPARABLE as possible.

• This is achieved by the use of SAMPLING STRATEGIES, rather than by exerting direct EXPERIMENTAL control over potential confounding variables.

• There has been NO RANDOM ASSIGNMENT to conditions.

• This kind of research is known as a QUASI-EXPERIMENT.

36

Statistical analysis

• Different statistical techniques are used for analysing the results of true experiments and the results of correlational studies.

• The results of a correlational study such as the one on the viewing of violent films are analysed with CORRELATIONS.

• The results of true experiments involve the COMPARISON of TYPICAL values, or AVERAGES.

37

Statistical analysis

• Like the results of a true experiment, the data from a quasi experiment are usually analysed by the statistics of COMPARISON (means, t-tests), rather than measures of ASSOCIATION (correlation coefficients).

• In the smoking study, the researcher would COMPARE smokers and non-smokers on their average age at death.

• We must always bear in mind, however, that in quasi-experiments, as in correlational studies, the supposedly crucial variable has been MEASURED IN THE PARTICIPANTS, not manipulated experimentally.

38

Potential confounds

• However many variables the researcher may try to control statistically, the smokers and non-smokers may yet differ on some crucial variable other than smoking.

• However many variables I control for, a critic can always argue that there may be some crucial respect in which smokers differ from non-smokers.

39

Evaluation

• Arguably, the quasi-experiment is essentially CORRELATIONAL, as opposed to EXPERIMENTAL research.

• All variables are observed as they occur in those studied.

• There is NO RANDOM ASSIGNMENT.

• Participants or subjects are SELF-SELECTING.

40

Research with animals

• Research with animals raises many problems, ethical as well as scientific.

• True experiments have been carried out, with random assignment and genuine experimental manipulation.

• These experiments show, unequivocally, that smoking is harmful – to certain animals.

• But animals are different from humans. I was once told by a veterinary surgeon that, to the domestic cat, aspirin is a deadly poison.

41

Experiments with human participants

• Psychologists study variables, not constants. • In this, psychology is no different from medicine,

biology, geography and many other subjects. • There is, however, an additional problem, which

psychology does not share with other disciplines.

• The human participant does not play a purely reactive role: The experimenter and the human participant INTERACT during the course of the experiment.

42

Demand characteristics(Orne, 1962)

• A DEMAND CHARACTERISTIC is a perceived requirement of an experiment.

• The participant will inevitably try to second-guess the experimenter’s purpose in running the experiment.

• When asked to memorise lists of words, the participant may use a mnemonic method to retain the lists and, as a result, achieve perfect recall of all the lists.

• But the researcher may have been trying to test the hypothesis that some kinds of lists are easier to remember than others.

43

Demand characteristics …

• The participants in a memory experiment will see the situation as a TEST, requiring (or ‘demanding’) that they perform as well as possible.

• In some universities, particularly in the US, students take many tests of aptitude and ability. Despite assurances of confidentiality, you would naturally want your records to bristle with alphas, rather than gammas, whatever you might be asked to do. Such a climate could be expected to encourage the use of mnemonics and other ‘tricks’.

44

The ‘experimenter effect’

• We owe this expression to Rosenthal (1966). • An experimenter’s expectations about how an

experiment should turn out may bias the recording of the results.

• An experimenter who believes that a participant has taken a performance-enhancing drug may observe improved performance when there is actually none.

• This is an UNCONSCIOUS bias, not deliberate cheating.

45

‘Bright’ rats, ‘dull’ rats

• An experimenter (see Rosenthal, 1966) tested two samples of rats on maze performance, having been informed that one sample was from an unusually ‘bright’ strain.

• Both samples had actually been drawn from the same strain.

• The ‘bright’ sample achieved higher scores.

46

Not fraud

• Rosenthal was NOT saying that the experimenter was guilty of cheating in any way.

• The recording of observations of behaviour (whether animal or human) often requires JUDGEMENTS about events which are by no means clear-cut.

47

Research with rats

• At one point, the white rat was much used in psychological research.

• As a measure of a rat’s ‘intelligence’, researchers were interested in the frequency with which an animal might cross the lines of a grid.

48

Uncertainty

• Of course, researchers try to agree on criteria for whether a rat has crossed a line or not.

• But I can assure you that, even armed with a detailed list of criteria, you often have to record Yes or No on an intuitive basis.

• In the circumstances, your expectations are likely to feed into your judgement.

49

Another research question

• Does the ingestion of a dose of caffeine improve performance on a skilled task?

• An investigator believes that it should.

• We shall try to design an experiment to test this hypothesis.

50

Performance on a skilled task

• The participants play a computer game that simulates shooting at a target.

• Each participant ‘shoots’ twenty times at a target and receives one point for each ‘hit’.

• The performance measure is the total number of hits by each participant, so each participant receives a score from 0 to 20.

• Number of Hits is the DEPENDENT VARIABLE in this experiment.

51

The need for a comparison

• The defining quality of a true experiment (as opposed to a DEMONSTRATION) is the possibility of a COMPARISON.

• There must be at least TWO conditions, one of which is a CONTROL condition, involving performance of the task WITHOUT ingestion of caffeine.

52

The independent variable

• The independent variable is Drug Treatment, one condition of which is the ingestion of a dose of caffeine.

• The other condition would be a comparison or CONTROL condition of some kind.

53

The comparison condition

• Could we simply take another group of volunteers and test them on the skilled task without giving them any caffeine?

• Mindful of the possibility of demand characteristics, we might wonder whether merely being given something to drink could affect performance, irrespective of what the drink contained.

54

Placebo effects

• A PLACEBO (Latin: I shall please) is a preparation with no pharmacological effects.

• A PLACEBO EFFECT is a kind of DEMAND CHARACTERISTIC.

• In the caffeine experiment, there must be a PLACEBO CONDITION. The control group would also ingest liquid (in the form of a drink, or intravenously).

• The placebo is usually a neutral saline solution, identical with the medium in which the drug dose was given.

55

Comparability

• The active and placebo conditions must be alike with respect to the manner in which the preparations are administered.

• Receiving an injection may itself have a greater effect upon performance than merely taking a drink.

• If an injection is given in the active condition, the Placebo group must also be injected.

56

The single blind

• The participant must be ignorant of (or ‘blind’ to) the true content of the drink. This is SINGLE-BLIND CONTROL.

• The single blind condition is intended to eliminate some of the DEMAND CHARACTERISTICS of the experiment.

• There remains, however, the possibility of an EXPERIMENTER EFFECT.

57

The double blind

• The DOUBLE BLIND is an experimental procedure in which neither the participant nor the experimenter knows the true purpose of the experiment.

• This device is intended to reduce both demand characteristics and experimenter effects.

58

Summary

• The psychological experiment is, in part, an INTERACTION between the experimenter and the participant.

• We must beware of DEMAND CHARACTERISTICS and EXPERIMENTER EFFECTS.

• This is achieved by incorporating the necessary CONTROLS in the experimental design.

59

Key terms

• at random

• event

• probability

• random assignment

• subjects and participants

• between subjects experiment

• within subjects experiment

60

Key terms …

• order effects

• counterbalancing

• cyclic permutation

• quasi-experiment

• demand characteristics

• experimenter effect

• placebo effect

61

Key terms…

• single blind

• double blind

62

A revision example