DESIGNING EXPERIMENTS

Psychologists aren't consistent in their terminology: I've seen this called "experimental design" and "participant design" as well as "research design". Edexcel in the past has used all three terms.

 “Design” refers to how the participants are assigned to different conditions of the IV in an experiment.

Usually an experiment will have a control condition who behave normally in a normal environment and an experimental condition who are different in some way – they might have had something done to them or their environment might have been changed.

• Some experiments have several different experimental conditions. For example, Baddeley (1966b) has two experimental groups (semantically-similar and acoustically-similar) and two control groups (the two dissimilar word groups); Bandura also has several experimental conditions compared to a control group who weren't exposed to any role model
  
• Some experiments don't have a control condition as such. In Milgram's obedience study, the original 1963 is the "baseline"; in Burger's repication of Milgram, Milgram's Variation #5 is the "baseline". The baseline isn't "normal" but it's what the experimental groups are being compared to.
  

Design decisions are about who gets to take part in each condition: the same people or different people?

INDEPENDENT GROUPS DESIGN

In an experiment with independent groups design, there are actually completely different groups of people in each condition of the IV and no one gets to be in more than one condition. This means that each participant only experiences the IV once.

This might be done when the researcher gets a sample together then splits them into two groups – perhaps by tossing a coin. One group becomes the experimental condition and the other group becomes the control condition.

• Sometimes a researcher has to use independent groups design. For example, if the IV is naturally-occurring, like “gender", and it's just not possible to assign someone to both conditions.

Independent groups produces confounding variables because it’s very difficult to make sure that the two groups contain exactly similar people. It’s hard to be sure that the IV is the only difference between the groups. These differences are called participant variables.

Participant variables are the main problem for independent groups design, because if the different groups aren't the same then the internal validity of the experiment is ruined.

Participant variables are always extraneous variables but they're not always confounding variables. Some participant variables just don't matter much. For example, in a memory experiment it probably doesn't matter what sign of the Zodiac each participant is; it's not like Leos or Capricorns have better memories. You don't need participants to be exactly identical, just similar in important respects.

Another (lesser) problem is convenience. You need a lot more participants to carry out independent groups design: a complete set of people for each condition. That can be time-consuming and (if you are paying them to take part) expensive.

REPEATED MEASURES DESIGN

Another approach is to take a single group of people and make sure they all experience both conditions. In repeated measures design everyone gets to be in the experimental condition and the control condition. In other words, they experience every condition of the IV.

This usually happens when you test people twice, at different times. For example, you could test students’ concentration levels before lunch then again after lunch.

In many ways, repeated measures is the best sort of design. It uses fewer people (since you don’t need two separate groups) and – crucially – you know that the only difference between the experimental condition and the control condition is the IV, because they’re the same people. To put it another way, there are fewer participant variables.

However, repeated measures can suffer from order effects. If you test the same group of people twice, then they might behave differently the second time because they’re familiar with the test – or bored with it.

COMPARING INDEPENDENT GROUPS & REPEATED MEASURES

If you are paying attention, you'll have noticed that independent groups and repeated measures are opposites of each other: the strengths of one are the weaknesses of the other, and vice versa.

ORDER EFFECTS, RANDOMISATION & COUNTER-BALACING

The big disadvantage of repeated measures is order effects: a confounding variable introduced because the participants experience every condition of the IV.

• Fatigue effects are order effects where the participants' performance goes down through boredom or exhaustion
• Practice effects are order effects where the participants' performance goes up through familiarity with the test

As well as practice and/or fatigue, order effects also make it much more likely the participants will figure out the purpose of the experiment and that introduces demand characteristics: the participants behave unnaturally because they are trying to do what they think the researchers want them to do.

You could get round order effects by just changing the test a bit each time. For example, you could use new words in a memory test. However, this means there are now situational variables at work and changes in the DV might not be solely due to the IV (the new words might be easier or harder to remember). This ruins the internal validity of the experiment.

There are two better solutions to the problem of order effects:

• Randomisation means determining which condition a participant experiences by random chance (tossing a coin, drawing a card). Some people might do the experimental condition first, then the control condition; other people might do it the other way round.
• Counter-balancing is similar, but there's no randomness. Just split the group into sub-groups: one sub-group does the experimental condition first, then the control; the other sub-group does it the other way round.
  

Both these techniques reduce order effects because although some participants do better (or worse) by doing the control condition second, others do the experimental condition second and they cancel each other out.

However, you need a lot more participants in your study if you're going to split them into sub-groups like this and one of the advantages of repeated measures design was supposed to be that it didn't involve using as many participants as independent groups design.

MATCHED PAIRS DESIGN

There is a "best of both worlds" option that combines the strengths of independent groups design (no order effects!) with the strengths of repeated measures (reduced participant variables!). This is matched pairs design.

Matched pairs starts out like regular independent groups design. Then you to “match” each person in the experimental condition with a similar person in the control condition. You could match people on their age, sex, IQ, social class or anything else that’s measurable.

The "matching" reduces participant variables if it is successful. However, there are problems with this:

• It's not always possible to find a participant who is a good match, especially if the experiment looks at unusual cases (eg Schmolck et al.'s study of brain damaged patients)
• It's not always possible to know in advance what variables you have to match people on (eg Raine et al. match participants on schizophrenia, but not on other mental conditions like psychopathy)
• Even if you can identify the variables you want to match participants on, it's not always possible to operationalise them (eg Schmolck doesn't seem to have matched "educational level" very successfully, because it still made a difference to the results, causing some of the MTL patients to outperform the control group)

APPLYING RESEARCH DESIGN IN PSYCHOLOGY
AO2 AND AO3 

Research design is a huge influence on validity, especially internal validity. The choice of design may increase the validity or decrease it, depending on the topic being studied and the procedures being used. Sometimes there are practical factors that force a researcher to use a particular design; if you want to compare criminals with non-criminals, it pretty much has to be independent groups design because you're either one or the other; if you want to compare reactions before and after taking a drug, you have to use repeated measures instead.

INDEPENDENT GROUPS DESIGN

Baddeley (1966b) is a good example of independent groups design. You can see the classic problem of large samples: he had to study 62 participants, just to get 15-20 taking part in each condition.

You can also imagine the major problem: participant variables. What if, through sheer bad luck, one of Baddeley's groups just happened to include people with really poor memories? or really good ones? When the semantically-similar conditin scored lower than the other coditions, this might be nothing to do with semantic LTM and entirely down to the rubbish memories of the people in that group.

However, Baddeley avoids order effects. His experiment was already tedious enough for participants (they got tested 4 times and then had a "forgetting test" sprung on them as a surprise); if he'd made them try the other conditions too, they would have become very demotivated.

Milgram's variations are also examples of independent groups, with the original 1961 study counting as "baseline measures". Milgram had to make sure none of the teachers had taken part before because that would have ruined his study: they would have realised the electric shocks were fake if they heard the same screams a second time.

However, Milgram ended up testing 780 participants and had to pay them all as well as putting hundreds of hours into the research.

Participant variables may have affected his results too. For example, in the early '60s fewer people had a good understandig of electricity, but those who did would not have been fooled by the experimenter's statement that the shocks "caused no permanent damage". Education and an understanding of the danger of electricity might explain why some participants rebelled, rather than the surroundings or the behaviour of a rebellious peer.

REPEATED MEASURES DESIGN

Watson & Rayner (1920) is a good example of repeated measures: Little Albert experienced both conditions, before and after the conditioning.

Because Albert was content to be around the white rat before the conditioning, we can be pretty sure it was the conditioning that caused him to start behaving in a fearful way. There are no participant variables at work.

Despite this being repeated measures, order effects aren't a huge problem. Usually, the more a child is around something, the more familiar it becomes and the less distressing it becomes. Conditioning produced the opposite effect in Albert. Therefore, the conditioning was so strong it overpowered any order effects there might have been.

Baddeley (1966b) also used repeated measures, because participants experienced the 4 recall trials then they experienced the forgetting test. Possibly, order effects did interfere with Baddeley's results. Both of the control groups showed a drop in scores on the forgetting test, but this might be because of fatigue ("Oh no, not again!?!") when the test was sprung on them rather than actual forgetting.

However, Baddeley couldn't have used randomisation or counterbalancing because the forgetting test had to come after the memory trials: you couldn't change the order around in this study.

MATCHED PAIRS DESIGN

You have several examples of these.

Schmolck et al. (2002) matched the brain damaged patients to healthy controls on two variables: age and education. However, when the MTL patients fractionally outperformed the controls on the tests of semantic LTM, Schmolck suggests this is because they were better educated. Clearly, this matching wasn't done carefully enough.

Bandura's Bobo Doll studies matched the children on aggression, from scores provided by the nursery teachers. This helped ensure that one of the conditions (say, the aggressive male model) didn't show the results it did because it just happened to have all the most aggressive children in it.

Raine et al. (1997) matched the murderers pleading NGRI to a group of controls based on age, sex and schizophrenia. It is interesting that Raine chose to match participants on schizophrenia but not on other mental health or other personality issues (eg psychopathy, empathy, depression). However, Raine didn't pair his participants against each other when he analysed his results, so his design is still independent groups despite his efforts to make the groups as similar as possible.

Becker et al. (2002) matched the Fijian schoolgirls on age and drew them from the same school. Some critics say Becker should have tried to match the girls on more variables, looking for girls in 1998 who were as similar as possible to the girls from 1995. After all, are you to-all-intents-and-purposes "the same" as students who were doing A-Levels three years before you?

EXEMPLAR ESSAY
How to write a 8-mark answer

Assess how research design is used in the Cognitive Approach. (8 marks)

• A 8-mark “apply” question awards 4 marks for describing the use of research design (AO1) and 4 marks for applying the Cognitive Approach to this (AO2). You need a conclusion to get a mark in the top band (7-8 marks). "Design" turns up in the Specification for the Cognitive Approach but you need to know how research design is used in all the Approaches.
  

Description
The Cognitive approach uses independent measures, which is which participants experience just one condition of the IV, and repeated measures, in which they experience all conditions of the IV.
There is also matched pairs design, in which participants experience just one condition but are matched on important variables.
Independent groups design removes the problem of order effects and repeated measures removes the problem of participant variables.
Matched pairs design also removes order effects and tries to reduce the impact of participant variables.

Application
Baddeley's memory study uses independent measures design by giving some participants acoustically-similar words to learn, some semantically-similar words to learn and a control group a list of unrelated words.
This forces Baddeley to use a very large sample of 62, split over 4 conditions, which is inconvenient and expensive.
Schmock uses matched pairs by matching the control group and the MTL group on education and age.
This was supposed to remove participant variables but Schmolck still blames the superior performance of the MTL patients on superior education.

Conclusion
The Cognitive Approach uses all types of design in its experiments but it does show how difficult it is to match participants on cognitive variables that are hard to operationalise, like "memory" or "education".

• Notice that for a 8-mark answer you don’t have to include everything about hypotheses. I haven’t mentioned order effects or counterbalancing. But it is a balanced answer - half description, half application.