Stats test Flashcards
Table
Spearmen’s Rho- IMD, Correlation, Ordinal
Mann Whitney U- IMD, Difference, Ordinal
Wilcoxon- RMD, Difference, Ordinal
Chi Squared- IMD, Difference, Nominal
Binomial Sign Test- RMD, Difference, Nominal
Difference- Diff between 2 IVs
Correlation- Link between 2 variables
Mann Whitney U test
1.Rank scores in ascending order regardless of group each score is in.
2. Lowest score gets rank of 1, if some data values are the same then use an avg rank e.g. 5.5
3. Ua= NaNb+ na(na+1)/2 -Ra
Ub= NaNb+ nb(nb+1)/2 -Rb
Na- no. of P’s in group A
Nb- no. of P’s in group B
Ra- sum of ranks for scores in group A
Rb- sum of rank for scores in group B
4. Calculate observed value- the smallest from value from Ua and Ub
5. Compare to critical value (from table) →then ↓
6. If OV is less than CV- test is significant as there’s a p less than 0.05 of EVs affecting results, if OV is higher than CV- test isn’t significant
7. Significant- reject null hypothesis and accept alternative hyp
Not significant- reject alternative hyp and accept the null
Spearman’s Rho test
1.Order data in ascending order with the smallest no. 1st. if there’s more than 1 of the same value you do the mean average
2. Calculate the difference between the 2 ranks/columns. Create row ‘d’
3. Square each result ‘d2’
4. Find sum of d2
5. Formula- 𝑟 = 1 − (6 ∑ 𝑑2) / 𝑛 ( 𝑛 − 1 )
∑ - sum of
n- no. of P’s
6. If Answer is +ve then it has has +ve correlation, if -ve then -ve correlation
Closer to 0 = weak correlation
Closer to 1 = strong correlation
7. Observed value has to be higher than critical value from table to reject null hyp and accept your hyp, but if it’s lower then accept null and reject yours
8. Choose the column for your type of hypothesis and value of p e.g. 0.05, read down until you get to the row matching your score for n- This is your critical value
Correlation- evaluation
✔
. Correlation can be studied on variables that can be measured but not manipulated
. Presence/absence of a relationship can be quickly seen, so need for further research is easily identified
✘
. Association doesn’t mean causation
. Correlstional analysis can only be done on variables that can be measured on a scale
Type 1 error
False positive
. When we say something is true but it’s not
. Accept hypothesis but result was due to chance
. p≤0.010- R’s 90% sure that results are not due to chance
type 2 error
False negative
.When we say something isn’t true but it is
. Accept null hypothesis but result was due to manipulation of IV
. p≤0.01- 99% confident in results
Proability symbols
≤- less than or equal to
≥- more than or equal to
≪- much less than
≫- much greater than
∼- an estimation
∝- proportional to
Wilcoxon test
Table:
1. Number of P’s
2. Score in condition 1
3. Score in condition 2
4. Calculate difference between C1 and C2
5. ‘Rank’ the differences in an ascending order, with smallest diff = rank 1, but if there’s more of same diff, use the mean value as their rank
6. Direction of difference- whether the value from the difference is + or -.
7. Add up the least frequent rank values e.g. if there are less -ves, add them- this is the OV
8. Find the CV from the table according to P’s, but remove P’s with result of 0 e.g. if there are 10 P’s but one of them has a diff of 0 there are now 9 P’s
9. If the OV is lower than/equal to the CV, results are signifcant, so reject null and accept alt.
Parametric tests- assumptions
- Population drawn from should be normally distributed
- Variances of populations should be approx. equal
- Should have at least interval or ratio data
- Should be no extreme scores
Use of non-parametric tests
- When assumptions of parametric tests aren’t met
- When distributions aren’t normal
