Lectures 1-3

Question 1

Define: experiment

Answer 1

Vary an independent variable while holding everything else constant to measure changes in a dependent variable. Therefore we can infer causality.

Question 2

Define: quasi-experiment

Answer 2

Independent variable cannot be manipulated (e,g, gender differences). DV is still measured as IV changes, but there is a problem with confounding variables.

Question 3

Define: correlational design

Answer 3

No manipulation, measure 2+ variables and determine to what extent they are co-related. Cannot infer causality.

Question 4

Define: nominal (categorical) scale

Answer 4

Mutually exclusive, not necessarily orders, categories. Calculations are meaningless. E.g. gender.

Question 5

Define: ordinal (ranking) scale

Answer 5

Numbers indicate a relative position (rank) in a list which is meaningful, although items are not necessarily equally spaced. E.g. questionnaire answer ranks, shoe size.

Question 6

Define: interval scale

Answer 6

The difference between two values is meaningful, but there is not a meaningful zero point. E.g. temperature in Celcius.

Question 7

Define: ratio scale

Answer 7

The difference between two values is meaningful and there is a meaningful zero point. Calculations (e.g. double) are meaningful. E.g. temperature in Kelvin.

Question 8

Define: confounding variable

Answer 8

A variable that confounds the interpretation of the results of an experiment. Some aspect of the experimental situation varies SYSTEMATICALLY with the IV. E.g. graphical stimulus was more interesting than verbal stimulus in the experiment.

Question 9

Define: nuisance variable

Answer 9

A variable that introduces noise but does not SYSTEMATICALLY bias the results. E.g. occasional noise during the experiment.

Question 10

Define: between-subjects design

Answer 10

Each condition is applied to a different group of subjects. Often the only available option depending on the IV (e.g. gender, task learning method). Individual differences between groups can be a problem, balancing this can be addressed by random group assignment.

Question 11

Define: within-subjects design

Answer 11

The same subject performs all levels of the IV. Also known as repeated measures design as they repeat the measure for each condition. Generally much more powerful, as each subject is their own control, eliminating individual differences. However, it can lead to a confound with order effects, e.g. the confound of practice or fatigue effects. This can be minimised by counterbalancing. However this can be complicated with complex designs.

Question 12

Define: matched-subjects design

Answer 12

Solves the problem of being unable to run within-subjects designs in certain cases - participants in each group are matched with a member of the other group and the data is treated like a regular within-subjects.

Question 13

Define: correlational design

Answer 13

Sometimes the test variables cannot be manipulated (e.g. for ethical reasons/time constraints) and instead pre-existing variables are measured in terms of the extent to which they are co-related or co-varying.

Question 14

Define: experimental hypothesis

Answer 14

Also known as a research hypothesis, it is a question addressed in an experiment, based on a more general theory.

Question 15

Define: statistical hypothesis

Answer 15

This involves precise statements about the data to be collected, e.g. explaining the IV and DV.

Question 16

Define: null hypothesis

Answer 16

H0, simply states that the different samples come from the same population. For parametric stats, often states that all the means are equal, for non-parametric that all the distributions are the same. It is the default, to be accepted unless there is good evidence to the contrary.

Question 17

Define: alternative hypothesis

Answer 17

H1/HA, the logical opposite of the null hypothesis - states that the conditions will have different means/distributions. The alternative and null hypotheses are therefore mutually exclusive and exhaustive.

Question 18

Define: theory, research hypothesis and statistical hypothesis.

Answer 18

A theory is a simple statement, e.g. ‘French lecturers are particularly great.’
A research hypothesis is more specific and testable, defining the IV, e.g. ‘The students of French lecturers perform better than the students of non-French lecturers’.
A statistical hypothesis defines the DV and states the null and alternative hypotheses.

Question 19

Define: test statistics

Answer 19

We reject the null hypothesis when p<α (usually 0.05). The test statistic is used in calculating p and has known probabilities associated with its values. It depends on the statistical test used.

Question 20

Define: p

Answer 20

p is the probability of collecting this data assuming that the null hypothesis is true, i.e. the probability that the result was found due to chance.

Question 21

Define: α

Answer 21

α is the criterion level set which p must be less than for the results to be believed to be more than just chance.

Question 22

Define: discrete data

Answer 22

Have certain fixed values, often integers.

Question 23

Define: continuous data

Answer 23

Can take any fractional value within given range.

Question 24

Ordinal scale variables are usually:

discrete/continuous?

Answer 24

Discrete

Question 25

Interval and ratio scales:

a) are usually continuous
b) are usually discrete
c) can be either continuous or discrete

Answer 25

Can be either continuous or discrete (e.g. time/no.of correct answers).

Question 26

The richest form of data representation is…?

Answer 26

Frequency distribution!

Question 27

In frequency distributions, representation of data depends on…?

Answer 27

Measurement scale, e.g. categorical, ordinal, ratio, interval.

Question 28

Categorical data can be presented as a frequency distribution with the y-axis being…?

Answer 28

Either n or the percentage of participants.

Question 29

Discrete variables can be presented as frequency distributions in various ways:

Answer 29

The frequency of each value (if the DV has only a few possible values), cumulative frequency, percentage frequency or cumulative percentages.

Question 30

It is often not possible when doing frequency distributions to calculate frequencies on the basis of every possible score. This can be solved by…

Answer 30

Using ranges or intervals of scores within which frequencies can be calculated.

Question 31

There is no particular rule for the number of ranges or intervals to use when doing a frequency range distribution, as it often depends on…

Answer 31

the number of samples.

Question 32

Usually with frequency ranges, …. intervals are used.

Answer 32

10-15

Question 33

Counting frequencies is a powerful technique for condensing data whilst retaining a great deal of information, but sometimes the entire frequency distribution shape is unnecessary and it is summarised further to a single number, a …

Answer 33

Measure of the central tendency of the data.

Question 34

There are three commonly used measures of central tendency - …

Answer 34

The mean, mode and median.

Question 35

Define mode and state the pros and cons of using the mode as a measure of central tendency.

Answer 35

The mode is the most common score or category in the data.
+ can be used for categorical data
+ always gives a real data value
- sometimes gives more than one value (bimodal distributions - two peaks?)
- varies depending on interval (bin) size - changing the interval sizes can change the mode.

Question 36

Define median and state the pros and cons of using the median as a measure of central tendency.

Answer 36

The middle value of the data set, or the mean of the two middle values if necessary.
+ insensitive to outlying data
+ often gives a real data value
+ intuitively appealing
- ignores a lot of the data
- difficult to calculate without a computer

Question 37

What kind of data does the mode tend to be used for?

Answer 37

Nominal.

Question 38

What kind of data is the median usually used for?

Answer 38

It’s the best measure for ordinal data and sometimes for skewed interval or ratio data.

Question 39

Define mean and state the pros and cons of using the mean as a measure of central tendency.

Answer 39

The mean is the sum divided by the number of samples: χ ̅=(∑x)/N
+ uses all the data and is therefore powerful
- very sensitive to outliers or skew
- does not always give a meaningful value
- only meaningful for ratio and interval data

Question 40

What kind of data is the mean the best measure for?

Answer 40

Ratio and interval data which is normally distributed.

Question 41

Outline measures of spread.

Answer 41

They are related to measures of central tendency - the median is associated with range or inter-quartile range (distance-based) and the mean with variance and standard deviation (centre-based).

Question 42

Define range and state the pros and cons of using it as a measure of spread.

Answer 42

Highest value - lowest value.

- very sensitive to outliers - the highest and lowest values are unlikely to be representative of the sample.

Question 43

Define interquartile range and state the pros and cons of using it as a measure of spread.

Answer 43

It is similar to the range but ignores the most extreme values - a quartile is the lowest score needed to include a given quarter of the population. The interquartile range is Q3-Q1 (the semi-interquartile range is the same divided by 2).
+ insensitive to outlying data
+ few assumptions about the data
- ignores a lot of the data
- difficult to calculate for large datasets without a computer

Question 44

Define variance and state the pros and cons of using it as a measure of spread.

Answer 44

Variance is: σ^2=(∑(x-x ̅^2)/N
\+ uses all the data, therefore powerful
\+ forms the basis of any other tests
- sensitive to outliers
- requires a normal distribution
- does not have a sensible unit

Question 45

Define standard deviation and state the pros and cons of using it as a measure of spread.

Answer 45

Standard deviation is the square root of the variance and therefore has a sensible unit. There are several variants of the formula, depending on whether a population (σ) or sample (s) is tested and whether this means to estimate for a population (ŝ). σ and s are just the square root of the variance equation, but ŝ uses N-1 as the denominator of the equation. This gives a better unbiased estimate of the population’s sd.