Features of Science
Science - a means of finding out about our world, i.e. gaining knowledge.
- Empiricism - information gained through direct observation or experiment rather than by a reasoned argument or unfounded beliefs.
- Objectivity - scientists strive to be objective in their observations and measurements, i.e. their expectations should not affect what they record.
- Replicability - repeating the observation to demonstrate validity.
- Control - in order to determine causal relationships, we must use the experimental method, i.e. IV and DV. All other conditions must be controlled.
The Scientific Process
- Theory Construction - using facts to construct theories. A theory is a collection of general principles that explain observations and facts.
- Inductive Model: Observations --> Hypothesis Developed --> Hypothesis Tested --> Data used to construct Theory.
- Deductive Model: Observations --> Propose Theory --> Hypothesis Developed --> Hypothesis Tested --> Draw Conclusions.
Psychology is a Science in that it shares the goals of all Sciences and the use of scientific method.
- Thomas Kuhn (1962): said Psychology could not be a Science because, unlike other sciences, there's no single paradigm (shared set of assumptions).
- It has been stated NOT a Science, as there is a lack of objectivity and control; problems like experimenter bias and demand characteristics.
- The Scientific approach is reductionist (because complex phenomena are reduced to simple variables) and deterministic (in its search for causal relationships).
Validating New Knowledge
Peer Review (aka refereeing) is the assessment of scientific work by others in the same field. Aims to increase quality of work in general.
- Reviewers are usually unpaid, and there are a number for each application/article/assessment.
- The Parliamentary Office of Science and Technology (2002) says peer review serves 3 main purposes: allocation of research funding, publication of research in scientific journals and books, assessing the research rating of university departments.
- Richard Smith (previous editor of British Medical Journal) said 'peer review is slow, expensive, profilgate of academic time, highly subjective, prone to bias and easily abused.'
- Unachieveable Ideal (poor research may be passed if appropriate expert is not designated), Anonymity (is usually practiced to promote honesty and objective, however reviewers may want to 'settle old scores' or bury rival research), Publication Bias (peer review tends to favour positive results), Preserving the Status Quo (peer review results in a preference for research that gos within an existing theory rather than unconventional research).
* Experiments (involve an IV and DV, all other variables are controlled so any changes to the DV are due to the IV NOT extraneous variables).
- controlled environment.
- high internal validity because extraneous variables can be controlled.
- control also increases replicability.
- however some things may reduce internal validity, e.g. investigator/experimenter effects and demand characteristics.
- reduced external validity because less like everyday life.
- natural experiment, more difficult to control extraneous variables.
- experimenter effects reduced because pps are not aware of being in a study.
- demand characteristics may still be present.
- makes use of existing IV, cannot manipulate IV so causal relationships cannot be determined.
- pps are not randomly allocated so may reduce validity.
- may be the only way to study certain behaviours or experiences.
Experimental Design - in any experiment there are several levels of the IV. Experimenters choose whether each pp is tested on all IVs (repeated measures), ot there are seperate groups for each IV (independent groups), or where the pps in each independent group can be matched with pps in the other group using key variables such as age and IQ (matched pairs).
* Self-Report Methods
Questionnaires and Interviews (finding out what people think and feel).
- can have an unstructured interview, where questions asked are based on previous answers by the interviewee.
- structured interviews/questionnaires are more easily repeated.
- may involve open questions where you can provide your own answer (rich insight but more difficult to analyse than closed questions).
- have to rely on honesty, e.g. social desirability bias.
* Cross-cultural Research
* Observational Studies - to watch what people do.
- use behavioural categories to record particular instances of behaviour.
- use sampling methods such as recording behaviour every 30 seconds (time sampling), or every time a certain behaviour occurs (event sampling).
- may have observer bias.
* Correlation Analysis - concerned with the relationship between 2 variables.
- doesn't establish a causal relationship, but can identify relationships.
- can be used with large sets of data and easily replicated.
* Case Studies - detailed study of an individual, institution or event.
- most are longitudinal, follow an individual or group over an extended period of time.
- difficult to generalise.
Implications of Sampling Strategies
Opportunity Sample - using pps that are most easily available, easiest method.
- biased because sample is drawn from a small part of the target population.
Volunteer Sample - can use advertisements etc.
- volunteer bias - pps are likely to be highly motivated and/or with extra time on their hands.
- location of advertisement could make it biased.
Random Sample - pps selected using random number techniques, 1st members of target population are identified, then pps are either drawn using lottery method or a random number generator.
- unbiased because all members of population have an equal chance of selection.
Stratified and Quota Sample - sub-groups (strata) within a population are identified (e.g. boy and girls or different age groups), then a pre-determined number of pps is taken from each sub-group in proportion to their representation in the target population. In stratified sampling this is done using random techniques, in quota sampling this is done using opportunity sampling.
- sub-group within sampling may be biased, e.g. because of opportunity sampling.
Snowball Sampling - directed to pps by existing pps.
- bias as researchers only in contact with a limited selection of the population.
Issues, Types, Assessing and Improving Reliability
- how much we can depend on any particular measurement.
- Experimental Research: reliability refers to its replicability.
- Observational Techniques: observations should be consistent; 2 or more observers should produce the same record. inter-rater or inter-observer reliability, the extent to which observers agree - total agreements / total number of observations, 0.8 is good inter-rater reliability.
- Self-report Techniques: internal reliability (being consistent within itself), external reliability (consistency over several occasions).
- reliability also concerns inter-interviewer reliability.
- assessing reliability: split half method, test-retest method.
Assessing and Improving Validity
- internal validity (whether the researcher tested what they intended to test).
- external validity (the extent to which the results of the study can be generalised to others, aka ecological validity).
- Experimental Research: artificial nature of lab. setting, however high external validity when setting isn't relevant to behaviour being observed, e.g. memory task. reduced internal validity as pps are aware they are in an experiment . lacks mundane realism.
- Observational Techniques: effected by observer bias. high ecological validity because they involve natural behaviours.
- Self-report Techniques: face validity (does the test look as if it's measuring what it intended to measure?), concurrent validity (established by comparing performance on a new questionnaire or test with a previously established test on the same topic). external validity is likely to be affected by the sampling strategy, which may create biased sample.
Ethical Considerations in Design and Conduct of Re
Ethical Issues with Human Participants
- Informed Consent and Deception - issues arise because full information may compromise the integrity of the study.
- Harm - what constitutes as too much harm? e.g. Ainsworth argued that distress in strange situation was no more than in everyday life.
Code of Conduct:
- Respect for the dignity and worth of all persons, (including privacy, confidentiality and informed consent. pps should be aware of right to withdraw.
- Competance - should maintain high standards in their work.
- Responsibility to their clients, public and to science. protection from harm, debriefing.
- Integrity - should be honest and accurate.
Dealing with Ethical Issues:
- ethical guidelines in the code of conduct.
- the use of ethical committees to assess research.
- educating students and qualified psychologists about their duties.
Ethical Issues with Non-Human Animals
Reasons for testing on non human animals:
- research may benefit animals.
- greater control and objectivity in research procedures.
- used when unable to use humans.
- humans and non-human animals have enough of their physiology and evolutionary past in common to justify conclusions drawn.
Is 'science at any cost' justifiable?
- sentient beings - do animals experience pain and emotions?
- speciesism - argued that discrimination of species is not different than racial or gender discrimination.
- animal rights - having rights comes with having responsibilities? or is it not acceptable?
- Animals Act (1986) - laws against animal research.
- The 3 R's - reduction (use of fewer animals), replacement (where possible use alternative methods), refinement (use improved techniques to reduce stress).
Probability and Significance
Inferential Statistics are about ruling out chance - allow you to make an educated guess about whether or not a hypothesis is correct.
- Instead of proving a hypothesis, you have to be content with finding out whether it's likely to be true - statistical significance (means they're unlikely to be down to chance, can read something into them).
- If results are not statistically significant, it means they could have happened by chance rather than being an effect of change of the IV.
Interpretation of Significance:
1. Write out null hypothesis (the theory you want to test).
- In a statistical test, you assume your null hypothesis is true (for the time being).
2. Choose a significance level ('level of proof' that you're looking for before you read anything into your results).
- The smaller the significance level, the stronger the evidence you're looking for that your results aren't just down to chance.
- A significance level is a probability, so a number between 0 (unlikely) and 1 (likely).
3. Turn all your experimental results into a single test statistic.
- Then you can find out how likely this test statistic is (and so how likely your results are), assuming the null hypothesis is true.
4. If the probability of getting your results is less than the significance level, then they must be very unlikely - so safe to say your null hypothesis isn't true after all.
- This is 'rejecting the null hypothesis', you assume your alternative hypothesis is true.
5. If you reject your null hypothesis, your results are statistically significant.
6. If you don't reject the null hypothesis, it means your results could have occurred by chance, rather than because your null hypothesis was wrong.
- If this happens, you've proved nothing - not rejecting the null doesn't mean it must be true.
7. Using a significance level of 0.05 (5%) is ok for most tests.
- If the probability of your results is less than this (p<0.05) then its good evidence that the null hypothesis wasn't true after all.
- If you use a significance level of 0.01 (1%), then you're looking for really strong evidence that the null hypothesis is untrue before you're going to reject it.
Types of potential error:
- Type 1 error - when you reject the null hypothesis when it was actually true.
- Type 2 error - when you don't reject the null hypothesis when it was actually false.
- If the level you choose is too big, you risk making a Type 1 error.
- If the level you choose is too small, you risk making a Type 2 error.
- One-tailed test - directional hypothesis.
- Two-tailed test - non-directional hypothesis.
Factors Affecting Choice of Statistical Test
- research may have either related measures (if a repeated measures or matched pps design was used), or unrelated measures (if an independent measures design was used.
- Some inferential statistics test whether there is a significant difference between 2 (or more) groups of scores: this is what happens in an experiment, IV is manipulated to see if it produces changes the DV that are significantly different from the control condition.
- Some inferential statistics test whether there is a significant association: this is what we look for in correlation studies, if they are associated, a significant correlation has been shown.
Levels of Measurement:
The results of a study can be collected in different ways, which affects how they can be analysed.
- Nominal Data - a frequency count for distinct categories.
- Ordinal Data - all of the measurements relate to the same variable, and measurements can be placed in ascending or descending rank order.
- Interval Data - measurements are taken on a scale where each unit is the same size, places pps in rank order according to the differences between them.
Spearman's Rho...is a correlation test.
- hypothesis predicts correlation between two variables.
- each person is measured on both variables.
- data is at least ordinal (i.e. not nominal).
Chi-Square...tests the null hypothesis, is usedwith nominal data and independent samples.
- hypothesis predicts differences between two conditions or association between 2 variables.
- data is independent.
- data in frequencies (normal).
- expected frequencies in each cell must not fall below 5.
Mann-Whitney U...is used with ordinal data.
- hypothesis predicts difference between 2 sets of data.
- independent groups design.
- data at least ordinal (i.e. not nominal).
Wilcoxon T...a test of difference for related data.
- hypothesis predicts difference between two sets of data.
- related design (repeated measured or matched pairs).
- data at least ordinal (i.e. not nominal).
Analysis and Interpretation of Qualitative Data
- quantitative methods are not relevant to 'real-life'.
- qualitative represents world as seen by individual.
- emphasises collection of subjective information from pp.
- data sets tend to be large.
- qualitative data cannot be reduced to numbers.
- can be examined for themes.
- quantitative: easy to analyse and produces neat conclusions BUT oversimplifies reality and human experience.
- qualitative: represents true complexities of behaviour, through rich detail of thoughts, feelings, etc. BUT more difficult to detect patterns and subject to bias of subjectivity.
Methods of Qualitative Data Analysis:
- Coding (the process of identifying categories, themes, phrases or keywords that may be found in any set of data) using top-down approach (thematic analysis) - code repesent ideas/themes from existing theory.
- Coding using bottom-up approach (grounded theory) - code emerges from data.
- Behavioural categories used to summarise data.
- Reflexivity indicates attitudes and biases of researcher.
- Validity demonstrated by triangulation (comparing results from a variety of different studies of the same things or person, the studies are likely to have used different methodologies, if the results agree, this supports their validity).
- Reliability checked by inter-rater reliability.
Central Tendency - indicates typical or 'average' score.
- Mean = sum of all scores divided by number of scores. unrepresentative if extreme scores.
- Median = middle value in ordered list of scores. not affected by extreme scores but not as sensitive as mean.
- Mode = most common value. not useful if there are many modes in a set of scores.
Measures of Dispersion - indicates spread of scores.
- Range = difference between highest and lowest score. not representative if extreme scores.
- Standard Deviation = spread of data around mean. precise measure but influence of extreme scores not taken into account.
- Bar Chart = illustration of frequency, height of bar represents frequency.
- Scattergram = illustration of correlation, suitable for correlational data. indicates strength of correlation and direction (positive or negative).
Reporting on Psychological Investigations
- Titles - should include IV and DV.
- Abstract - concise summary about research and findings.
- Introduction - general overview of aea being studied.
- Aim and Hypotheses - state purpose of study, and IV and DV included hypothesis.
- Method - Design of Investigation: research method, research design, how variables were controlled, how ethical issues were dealt with. Procedure Used: what happened, what was carried out, how data was recorded. Use of PPs: number of pps, sampling methods, how they were allocated. Resources Used: materials, apparatus.
- Results - results can be reported as descriptive or inferential statistics.
- Discussion - explanation of findings, implications of study, limitations and modifications, relationship to background research, suggestions for further research.