Designing an Investigation
The following factors are important to consider when designing an investigation:
Target population: group that the R is interested in and from whom the sample is drawn and generalisations can be made.
Pilot study: Is a test run on a few P’s enabling you to check for design faults and to see if there could be any improvements to the study before carrying out investigation on a large scale, this is a routine procedure.
Confederate: Person (not a P) assigned by the experimenter to behave in a certain way to affect the experiment. May be used as IV.
(Correlation is not a R method so don’t say a correlational study but a correlational analysis)
Correlation/correlational analysis: Determines the extent of a relationship between 2 variables. Usually a linear correlation is predicted but can be curvilinear relationship (i.e. Yerkes-Dodson law.)
Positive correlation: 2 variables increase together
Negative correlation: As one variable increases, the other decreases; the tighter the points cluster around the single straight line the
Zero correlation: No relationship (Casual relationships can be ruled out if not correlation exists)
Correlation doesn't mean Causal
Correlation doesn’t mean that one variable caused the other to change and only an experiment reveals causal relationships between variables. Causality is only one of three possible explanations for a correlation
1. Relationships is causal (1 variable caused the other to change)
2. The relationship is due to chance (2 variables just happen to be statistically related).
3. There is a third factor involved (another variable is causing the relationship).
So casual relationships can be ruled out if no correlation exists
Because 2 variables rise and fall together doesn’t mean they cause each other e.g. intelligence increases with age (intelligence and age are the co-variables) but age doesn’t cause intelligence.
Correlational Analysis - +'s and -'s
+ Used when unethical/impractical to manipulate variables / They can indicate trends leading to further R.
+ Allows a R to measure relationships between naturally occurring variables e.g. height and intelligence.
+ If correlation is significant then further R is justified and if not you can rule out causal relationship.
+ As with experiments the procedures can be repeated again and findings can be confirmed.
- People often misinterpret correlations and assume that a cause and effect have been found, not possible to draw conclusions about case and effect.
- Coefficient may look like there is no relationship between V's as it is near to 0 but it may be hiding a curvilinear relationship or one which shows more than one group in the data. If you calculated the correlation coefficient for a curvilinear relationship - find something close to 0 as half the time the relationship is + the rest of the time it is - and together they cancel each other out and get a 0 correlation
- There may be other unknown variables (intervening variables) which can explain link between co-variables.
- As with experiments may lack internal/external validity e.g. method used to measure IQ may lack validity or sample used may lack generalisability.
Correlational Analysis - Scattergraphs
Graphical presentation of the relationship (correlation) between 2 sets of score’s.
The scatter of the dots indicates the degree of correlation between the co-variables.
Drawing a scattergraph includes plotting 2 scores: one score is measured along the horizontal axis whilst the other is along the vertical axis and when the 2 plots intersect on the graph an X plotting point is placed. If doesn’t show type of correlation clearly draw a line of best fit (doesn’t have to pass through any particular number of x’s) unless need to draw a line of bets fit just leave it out. The pattern of points plotted on the scattergraph represents particular types of correlation.
Hypothesis - Correlation
When conductioning a study using a correlational analyssis. You need to produce a correlational hypothesis, which states the expected relationship between co-variables
Age and Beauty are the co-variables , the study expects to find a relationship bteween these co-varaibles, so possible hypothesis might be.
- Age and Beauty are positively correlated (Directional)
- As people get older they are rated as more beautiful (Directional)
- Age and beauty are corrleated (Non - Directional)
Observations - Naturalistic
Observation: Systematically watching and recording what people say and do i.e. how they behave.
Data gathered through observation is highly descriptive and will not offer an observation for what has been recorded. It’s the job of the R to make sense of the data, sorting it so that any evidence relevant to the hypothesis is presented clearly.
Naturalistic: A research method in a naturalistic setting where the investigator doesn’t interfere but observes the behaviour in question, though this is likely to involve the use of structured observation. Before starting the study, observers try to become familiar to whom they’re observing, to minimise the effect their presence has.
Observations (Naturalistic) - +'s and -'s
+ Few demand characteristics as P’s don't know they're being studied and not in a false situation=higher internal validity compared to questionnaires/interviews as what P's say they do if diff to what they actually do.
+ Info collected more detailed/provides a fuller pic of behaviour than the info collected in a laboratory.
+ High ecological validity since the behaviour occurs in its true form in a natural setting.
+ Can be used when other methods not possible e.g. might be unethical/ P’s unwilling to fill in questionnaire.
- Risk of observer bias as unlikely that R can remain completely objective, reduces reliability of data gathered.
- Control of environment not poss and confounding V's introduced near imposs determine causal relationships
- Replication would be difficult.
- Ethics are big prob. esp. naturalistic. Not knowing being watched e.g.1 -way mirrors issues with privacy, informed consent and confidentiality.
- Tend to be small scale so group studied may not be representative of the population - lack pop. validity.
- Lots planning - choosing V's to operationalise, creating behavioural categories and devise recording m
Observations - Controlled
Controlled: Observations that take place where some variables in the P environment are controlled and manipulated by the experimenter
Often used in field experiments as an IV is being tested and control is possible. In order that control can be exerted and so that the behaviour is easier to observe. E.g. watch aggressive film or not
+ By controlling some variables, it is possible for the R’s to draw conclusions from their observation and is also easier to establish cause and effect.
- An unfamiliar setting may affect participant’s behaviour, making it less natural (ecological validity)
- P’s may be aware of being observed creating demand characteristics
Observations - Structured or Unstructured
Structured (systematic) observation: An observer uses various ‘systems’ to organise observations such as behavioural categories and sampling procedures.
+Gather relevant data as you know what you are looking for
-Intresting behaviour could go unnoticed as you are not looking for it
Unstructured observations: An observer records all relevant behaviour but has no system. This technique may be chosen as behaviour to be studied is likely to be unpredictable.
Observations (Structured) - Behavioural Categories
Need to operationalise the observation/behaviour to create behavioural categories i.e. a set of components
Behavioural categories: Dividing target behaviour in to a subset of behaviours. This can be done using a behaviour checklist or coding system. This improves reliability.
Behavioural Categories should:
- Be objective - record actions don’t make inferences
- Cover all possible component behaviours – don’t cover things with are not necessary
- Be mutually exclusive exclusive – don’t mark 2 categories at once i.e. hitting and shoving
Observations (Structured) - Sampling Procedures
Sampling Procedures: If conducting continuous observation R records every instance of behaviour in detail. In many situations not possible as it creates too much data.
So could use
Event sampling: An observational technique in which a count is kept of the number of times certain behaviour (event) occurs.
Time sampling: An observational technique in which the observer records behaviours in a given time frame e.g. every 30 seconds you may select more than 1 category from a checklist.
Observations - P and Non-P
Non-Participant: the experimenter does not become part of the group being observed
+ R can remain objective throughout
- The R loses a sense of the group dynamics by staying separate
Participant: Observer becomes one of the groups of P’s he wishes to observe. Observer may tell the others they will be observed (an overt observation), or may pretend to be one of the group and not inform them that they are being observed (a covert observation).
+ Can observe P's in a natural setting (high ecol validity) and gain understanding of causes of their behaviour.
+ The R develops a relationship with group - gain a greater understanding of the groups behaviour
- Remembering accurately may be difficult as unable to take notes.
- Observer loses objectivity - may interpret or record information in a biased way.
- P’ may act differently if they know a R is amongst them
- Ethical guidelines such as deception, consent and confidentiality may not be maintained.
Observations - Overt or Covert
Overt: people being observed do not know they're being watched or studied. But knowing that behaviour is being observed is likely to alter P's behaviour.
Covert/Undiscolsed Observation: Participants are unaware they are being watched e.g. one way mirror.
Observers try to be as unobtrusive as possible (to minimise hawthorne effect) though this is has ethical implications
Designing Observational Research
Are you using observation as a method or a technique?
Controlled or Naturalistic?
Overt or Covert?
Structured or Unstructured?
If Structured - what sampling procedures/ behavioural categories and methods to observe i.e. behavioural checklist, coding system or rating system?
Coding system: Systematic method for recording observations in which individual behaviours are given a code for ease of recording e.g. PLYO – playing when with owner.
Behaviour Checklist: A list of the behaviours to be recorded during an observational study
Rating System: E.g. Early Child Environment Rating Scale. Records observations of child’s early environment and rates items on a 7 point scale (1 is inadequate and 7 is excellent.) This is then related to other developmental outcomes such as school success.
Evaluating Observational Research - Validity
External Validity - Likely to be high as they involve more natural behaviour
Population Validity - May be a problem e.g. if children are only observed in middle class homes so we can't
Internal Validity - Observations will not be valid if the coding system is flawed
Observer bias: if what someone observes is affected by the expectations (they may see what they expect to see) This reduces the objectivity and validity of the R 2 or more observers make this worse. Can check observer reliability by inter-rater reliability and conduct pilot study.
Improving validity: Carry out R in varied setting with varied P’s and use more than 1 observer to reduce observer bias and average data across observer (to balance out any biases)
Ethical issues: This type of R is acceptable where those observed would expect to be observed by strangers. However R’s should be aware that’s not acceptable to intrude upon privacy of individuals who even whilst in public space may believe they are unobserved.
- In studies where P's are observed without their knowledge there are issues relating to informed consent.
- Observations - invasion of privacy (1 way mirrors is deception) so P confidentiality should be respect
Evaluating Observational Research - Reliability
Refers to whether sonething is consistent . So any tool used to measure e.g. observations or interviews must be reliable (so prodce and so should produce the same result on every ocassion - if it doesn't must check the thing has changed and not our measuring tool.
Reliability of Observations
Inter-rater reliability: Extent to which 2 observers agree. To ensure reliability, have at least two observerswatching and recording what they see in the same way. Judges will often score observations in categories and the % of agreement between the judges will be calculated. E.g. if judges score observations in same category 8 times out of 10, there’s an 80% inter-rater reliability rate. If total agreements/total observation over 80% = reliable.
Observers must be trained in the use of coding systems/behaviour check list and thet must prcatise using them and discuss their observations. The investigator can then check how reliable they are.
Evaluating Observational Research
You can have a study that is reliable but lacks validity
E.g. if an observer uses a behaviour checklist which is not very thourough and sometimes the target individual does things which can't be recorded, the observation may be prefectlt reliable but lack validity because the behaviour checklist was poor.
Self-report - Questionnaire +'s and -'s
Questionnaire: Data is collected through use of written questions.
+ Can be easily repeated so data can be collected from large no’s of people quickly, cheaply and easily so is efficient.
+ P’s are anonymous, so are more willing to reveal personal info/more truthful than in interview, reliable method of gathering data.
+ People who are geographically distant can be studied.
- Answers may not be truthful because of leading q’s and social desirability bias or may just deliberately give the wrong answers.
- Difficult to obtain a representative sample as it is difficult to identity all members of a population and there is no guarantee that all will agree to take part in the study. It may be that those who do agree make up a biased same because only certain types of people fill in questionnaires e.g. literate people who are willing to spend time filling it in and returning it.
- Survey data is highly descriptive as so it’s difficult to establish causal relationships and the ability to infer causal relationships will be limited by the quality of the questionnaire.