Research Methods: Study Designs

• These are research methods/statistics study notes

Research Design: Hypotheses

• The null hypothesis predicts no relationship between variables.
• The alternate hypothesis predicts a relationship between variables.

Designs of Psychological Studies

• Experiments involve manipulating a variable in a controlled environment to measure changes and infer a causal relationship.
• Correlational studies measure two variables to see if there's a relationship, without manipulation, so causation cannot be inferred.
• Quasi-experiments occur when people organize themselves into a kind of experiment, without researcher intervention.

Groups

• Independent groups: Participants are allocated to different groups to be measured
• Repeated measures: All participants are in the same group and receive the same manipulation.
• Counterbalanced groups: Reduces practice and fatigue effects by changing the order in which participants perform tasks.
• Matched participants: Reduces the impact of individual differences.

Variables

• Experiments have at least two variables:
  • Independent (manipulated)
  • Dependent (measured)
• Correlational studies also have two variables, but no independent variable because there is no manipulation.
• Operationalising variables: Putting variables into a form that can be directly measured

Psychology Statistics: Central Tendency

• A measure of central tendency gives an indication of what a typical mid-point and central score might mean

Three Measures of Central Tendency

• Mean: Average.
• Median: Middle score.
• Mode: Most common score (can have more than one)

Mean: Advantages & Disadvantages

• Advantages: Takes all scores into account, reflects the whole sample.
• Advantages: Mode and median may not change, but the mean will if scores are added to sample
• Disadvantages: Impacted by outliers

Median: Advantages & Disadvantages

• Advantages: Less sensitive than the mean, less likely to be distorted by others

Mode: Advantages & Disadvantages

• Advantages: Useful when the typical scores are not the ones in the middle
• Disadvantages: Not useful when there's a very large scale where scores are not likely to occur more than once

Describing a Dataset: Measure Sensitivity

• Mean is best because it is the most sensitive
• Median is next best because it is not impacted by outliers, so long as typical scores are towards the middle of sample
• Mode is next best, so long as the sample has scores that occur multiple times.

Measures of Dispersion: Range

• Range measures the difference between the highest and lowest scores in a sample
• Range does not give an indication of where most scores are (at the top or bottom).

Interquartile Range

• The distance between the bottom 25% and top 75% of scores.
• Useful as it gives an idea of where the bulk of the scores lie, discounts any unusually high or unusually low scores.

Standard Deviation

• The average amount of scores in a data set that vary from the mean.
• Higher SD scores mean the data is more spread out (dispersed).
• Lower SD scores mean the data is less dispersed (clusters together around the mean more).
• Can be used to interpret how high or low an individual score is relative to other scores

Variance

• The SD squared (SD is square root of the variance of a dataset)
• Bigger numbers mean more variation in the data.

Standard Error of the Mean

• An estimate of how likely it is that your dataset represents the whole population
• SEM is derived by dividing the SD by the square root of the number of scores in your sample
• Larger SEM scores are less likely to represent the overall population, while smaller SEM scores are more likely
• Larger dataset + smaller level of dispersion: data is more reliable (smaller SEM)
• Smaller dataset + data is all over the place: data is less reliable (larger SEM)

Dispersion

• SD, Variance and SEM gives more information about the dispersion of the dataset
• Can be misleading if data is not normally distributed (range & IQR are more useful)

Inferential Statistics

• Statistics that estimate or infer something from our data which we can't directly observe

Inferential Statistics: Rationale

• We need inferential statistics to estimate things like:
  • How likely it is our data represents the overall population.
  • How likely it is that any difference or relationships in our data occurred due to random chance.
  • How likely it is that any differences or relationships in our data occurred due to a relationship between our variables that actually exists in reality.

The P-Value

• An estimate of how likely it is that the observed result could have occurred if the null hypothesis were true.
• If there is a low likelihood that the data could have occurred if the null hypothesis were true, that is an argument for rejecting the null.
• A p-value is a percentage score, expressed as a decimal.
• If p = 1, that means there is a 100% chance of this data occurring if the null is true
• Lower numbers = less likely that the result would have occurred (higher likelihood that alternate hypothesis is true)

Statistical Significance

• If your p-value is <0.05, you can reject the null hypothesis
• Demonstrates that the variable are related - manipulating the IV really does impact the DV (relationship between the two)

Draw Conclusions

• p tells you whether any differences in the data occurred due to random chance or due to an effect that really exists
• p doesn't tell you what the relationship between your variables actually is; just whether the relationship happened due to change or a real effect

Correlation

• Researchers look for correlations and are interested in the strength and direction of the relationship

Relationships: Direction

• Positive correlation: when one variable goes up, the other variable also goes up
• Negative correlation: when one variable goes up, the other goes down (sometimes called an ‘inverse correlation')

Relationships: Strength

• Variables can be strongly or weakly correlated. This means that the level of one variable has a large impact on levels of the other variable, positive or negative
• Correlation coefficient (r) can be a score between 1 and -1:
  • 1 means perfect positive correlation (if x goes up by 1, y goes up by 1).
  • -1 means perfect negative correlation (if x goes up by 1, y goes down by 1).
  • 0 means no correlation at all (when x goes up/down by 1, y goes up/down by 1).
• You can judge the direction of the relationship by whether the coefficient is positive or negative
• You can judge the strength of the relationship by how far the r value is from 0 (further from 0 is stronger).
  • Closer to 0 is weaker
  • R values between 0.75 and 1 are strong relationships
  • R values between 0.5 and 0.75 are moderate
  • R values between 0.25 and 0.5 are weak
  • R values between 0 and 0.25 have no correlation

Finding P

• r = measures strength and direction
• You need p to estimate whether that relationship is likely to exist in reality or whether it occurred due to random chance
• Higher r results result in a lower p. A strong relationship is more likely it exists in reality

Types of Data

• Four types: Nominal, Ordinal, Interval, Ratio

Nominal Data

• Data that only allows us to sort scores into different categories
• Cannot be arranged in an order and cannot make a numerical comparison between them.

Ordinal Data

• Numerical data that allows us to arrange scores in a numerical order, does not allow any other kind of numerical comparison.
• Doesn't tell us the distance between scores, only the order they can be arranged in

Interval Data

• Similar to ordinal data, allows us to rank scores from high to low (allows us to judge the distance between scores).

Ratio Data

• Similar to interval data, except that the scale has a real zero point, which means that the bottom of the scale represents an absolute absence of the thing being measured.
• Absolute zero point means you can make proportional judgements about your data points.