Types of Research Deign¶
There are three main types of research:
- High risk, high impact
- Seeks to expand our understanding by testing phenomena
- In the name: it “explores” the topic in question to see what’s happening
- Seeks to catalogue and categorise the findings of exploratory research
- Will tell us what is going on, but not why
- Methodology may be quantitative or qualitative (sometimes both)
- Typically uses descriptive statistics to present correlation but not causation.
- Sometimes called “analytical research”
- Specifically asks the question of why and looks directly at causality
- Typically conducted through hypothesis testing to try and break assumptions
- Methodology is always quantitative and thus blanks and controls are vital
Data collection methods¶
Methods can be quantitative (structured approach), qualitative (unstructured approach), or a combination of both. This is a descriptor of the type of data that you’re working
|Uses||Surveys, polls and experiments||Interviews, observation and open-ended
forms of data collection
|Sample size is||Larger with better randomisation||Smaller with purposeful choice|
The following questions should be answered:
- Is the logic of the experiment sound?
- Can a blank be performed (negative control)?
- Can a control be performed (positive control)?
- Can standardisation and calibration be implemented?
- Have all likely factors and interferants been accounted for?
True experimental design¶
Is characterised by random assignments to treatments
- Participants are randomly assigned to groups
- The participants are observed for baseline measures (pretest)
- One group is given the treatment and the other is given a control
- The participants are observed for changes to the baseline (protest)
Quasi experimental design¶
Meets some, but not all of the experimental criteria (e.g. no pretest, subject assignment may not be random). This may be necessary if the research project is looking for a particular demographic or condition within their subjects.
Is querying a population rather then testing them. The results come from inference rather than explicit testing. E.g. testing smokers. This is done by using data from surveys (from a sample of the population) and census (the whole population)
Comes in different types of studies
Casal comparative studies¶
Borrows strategies from experimental design to answer causal questions. Instead of assigning groups to control and experimental paths, they recruit people of similar demographics who differ by the causal relationship in question.
E.g. recruit a whole bunch of people, half of whom are smokers.
This type of study is very prone to confounding variables and may not have a close enough demographic match to compare the two groups.
looks at one sample at a single point in time. These are the weakest form of study, but allow us to look at rare cases and situations in great detail.
These studies are not generalisable to the greater population.
Look at multiple samples at a single point in time. The process is that a population is chosen and is separated into those with and without a predisposition. Each of the two groups is then broken down into those who have been exposed to a situation and those who haven’t.
Only really classifies as descriptive research
Examine the same people over a long period of time. Tests for an association between variables in the same entity at different points in time. This type of study is incredibly powerful but is also incredibly high investment.
Subjects may leave the study, so over recruiting is a must
Time-series study - longitudinal¶
Looks at one sample or group at multiple points in time. A type of longitudinal study that follows a cohort with changing individual subjects, rather than following the same specific people.
This study is not generalisable
Panel study - longitudinal¶
Looks at multiple samples (not necessarily the same) over multiple points in time. The subjects are representative of a greater population and are followed over a long period of time. These studies may also follow subjects over multiple generations.
Cohort study - longitudinal¶
Follows multiple samples (the same ones) over a long period of time. Is similar to a panel study, but seeks to ask questions about the cohort. The category is analysed as a whole and the results are generalisable and reliable.
Randomised controlled trial¶
Participants are randomly allocated to treatment or control groups and are assessed for outcome.
Each participant receives both therapies, allowing for each subject to act as their own control. Statistical tests that assume randomisation can be used.
The study is only relevant if the treatment is non-permanent
Case control study¶
Examines in-depth, many features of a few cases over a duration of time. The data can be much more broad, detailed and varied, though qualitative data should be taken about a few cases. This is most useful for very rare disorders, or disorders that have a long delay between exposure and outcome. Fewer subjects are needed than for cross-sectional studies.
Factors to consider with experimental design¶
Is a measure of how true to the correct answer the outcome is. This comes in multiple forms:
- Internal validity - The cause/effect relationship that was found in the study is really there and cannot be explained by some other factor. Instrumentally, the machines should measure what they’re supposed to.
- External validity - The results obtained in the study are the same as can be seen in samples outside of the study. Can the results be generalised?
- Practical validity - Do the results have meaning outside of the study itself?
A measure of the consistency of results. The results should be consistent across time and across samples.
Sources of error¶
Error is the difference between the measured result and the true value
- Random error - The “noise” in the data. This is unavoidable and is often not able to be dealt with. More samples may account for random error in the experiment. This will show up as a component of the precision of the result.
- Systematic error - Is inherent to the experiment, equipment or methodology. It will typically present as an overall bias in the results, shifting the accuracy, but not the precision.
- Human Error - This is an outright mistake made by the experimenter. May be deleted data, incorrect use of equipment, incorrect enactment of the methodology or physical accident/clumsiness.
Will either intentionally or unintentionally influence the interpretation of the results. This is a type of systematic error. Bias is inherent to studies carried out by humans and should always be considered to be an influencing factor.
Criteria for causality¶
While correlation is not causation, correlation is the first check. If there is a real correlation, then the following three criteria must be met to determine causality:
- Temporal precedence - Does the theorised cause come before the effect?
- Covariation - When the theorised cause happens, does the effect always follow?
- Elimination of extraneous variables - Are there any other plausible causations? Easiest to prove this using covariation with strict variable control of all extraneous variables.