Statistics

Question 1

Correlations + eval

Answer 1

analyse strength and direction between co variables from -1 to +1
plot covariables on a scattergraph

+Establish strength between 2 variables and measure precisely
+Predictions can be made based on correlations
+Can investigate things that cannot be manipulated experimentally for ethical or practical reasons
- Cause and effect unclear
- Third variables - correlation and causation
- Doesnt detect curvilinear relationships - positive until a certain point then negative (e.g enzymes) or vice versa

Question 2

What do significance levels mean

Answer 2

Level at which decision made to reject null hypothesis, so how sure we are IV is acc having affect on DV and not by chance. Significance level = probability that the null hypothesis is accepted by chance

Question 3

Conventional significance levels for errors

Answer 3

Type 1 error = false positive = probability that null hypothesis is wrongly rejected = significance level
Type 2 error = false negative = probability that null hypothesis is wrongly accepted - harder to calculate as depends on strength of effects and sample size etc
5% commonly used as a middling fair value to minimise both errors, but sometimes 10% used for more lenient experiments, and sometimes 1% used for stricter topics like medicine (where type 1 errors cannot happen)

Question 4

Levels of Measurement

Answer 4

Nominal - discrete data in separate categories e.g eye colour, one person can only be in one category
Ordinal - continuous data with some sort of ordering or ranking e,g listing music genres in rank order, or using own scales e.g scale of 1-10 with arbitrary units
Interval - continuous data in equal intervals like height or time

Question 5

Parametric vs non parametric tests

Answer 5

Parametric tests more robust, rely on actual data rather than rank/categories and more likely to accurately detect significance
Require interval data, somewhat normally distributed population and similar variance of scores between conditions (so standard deviation is similar)

Question 6

Factors to choose a test

Answer 6

Whether it is correlation (spearman’s rho or pearsons r) association (chi squared) or difference (everything else)
Research design - independent measures (chi squared, mann whitney u, unrelated t) , repeated measures/matched participants (these 2 joined together)(sign, wilcoxon, related t)
Level of measurement - nominal (chi squared, sign test, chi squared), ordinal (mann whitney u, wilcoxon, spearman’s rho), interval/parametric (unrelated t, related t, pearsons r)
Use mnemonic Carrots Should Come Mashed With Swede Under Roast Potatoes

Question 7

Critical value

Answer 7

numerical value that helps determine significance of results by comparison to test stat - boundary value of something that doesn’t normally happen under the null hypothesis

Question 8

Spearman’s rho

Answer 8

relationship/correlation, ordinal data, related pairs, non parametric
Strength from -1 (-ive) to +1 (+ive correlation)
Test against crit value using modulus but then convert back to -ive if needed for description of results
If test stat more than equal to crit value then significant

Question 9

Mann Whitney

Answer 9

Difference, independent groups, ordinal data, non parametric
Formulas used for test value
If smaller or equal to critical value then significant

Question 10

Chi-squared

Answer 10

Difference/association, independent data, nominal, non-parametric
Same method as FM
Test stat must be greater or equal to critical value to be significant

Question 11

Wilcoxon

Answer 11

Difference, related pairs, ordinal, non-parametric
Test stat formulated
Must be less than or equal to critical value to be significant

Question 12

Sign test

Answer 12

Difference, related pairs, nominal, non-parametric
Make a hypothesis (1/2 tailed), then work out the sign change for each participant by doing experimental score - control score, then total up number of + and - where test statistic S is lower value
N is number of participants but ignore 0 scorers
Then match up critical value where S needs to be equal to or less than for it to be significant

Question 13

Pearson’s r

Answer 13

Correlation/relationship, related pairs, interval data, parametric test
Same method as fm PMCC
Test stat must be greater than or equal to critical value to be significant

Question 14

Related t-test

Answer 14

Difference, related pairs, interval, parametric
Test stat formulated
Crit value based on df = number of pairs - 1
must be greater or equal to critical value to be significant

Question 15

Unrelated t-test

Answer 15

Difference, independent groups, interval, parametric
Test stat formulated
Crit value based on df = total sample size of both groups - 2
must be greater than or equal to crit value to be significant