© Charfi 2021 © Charfi 2021

Review
• Perform bivariate statistics (associating scale
data: Pearson)
Inferential Statistics • Pearson (r) assumptions
– the two variables must be measured at an interval ratio level
.................................................................
– there is a linear relationship between the two variables
.................................................................
Lecture 5: Bivariate analysis 2 – there is no significant outliers
.................................................................
(Spearman and Chi2) – the two variables must be aproximitly normally distributed
.................................................................
LBA 2
• Reverse coding
A course designed and taught by
Dr. Hedi Charfi 1 2

© Charfi 2021 © Charfi 2021

Lecture objectives Test choice

Objectives: • Number of variables
• Perform bivariate statistics • Test objective
– (Associating scale and ordinal variables: Spearman)
– (Associating nominal variables: Chi2)
• Level of measurement
• Examine Spearman and Chi2 assumptions + Test assumptions
• Coding variables (introducing missing
values)
3 4
 © Charfi 2021 © Charfi 2021
The case of failure to meet Pearson 2. Spearman Rank-Order Correlation (Rho)
assumptions and associated measures (ordinal data)
• Correction • Spearman (Rho) measures the strength of a
monotonic relashionship
• Switching to non parametric tests (e.g ................................................................
Spearman) between two variables measured at the
ordinal or interval ratio level
................................................
• A value of zero (or close to zero) indicates
dissociation between the variables.
• The formula for Spearman Rho is:

(Gray & Kinnear, 2011)

5 6

© Charfi 2021 © Charfi 2021

Spearman Rho assumptions Checking for monotonicity

• Both variables are measured at
ordinal or interval ratio level
...........................................................

• There should be a ...........................

monotonic
relationship between the two variables
(otherwise, Eta should be run)
linear and monotonic monotonic non linear
For smaller samples use Kendall Tau  more robust non monotonic
pearson spearman
Eta
(Gray & Kinnear, 2011)7 8
 © Charfi 2021 © Charfi 2021

Checking monotonicity in SPSS

9 10

© Charfi 2021 © Charfi 2021

Interpreting output (same steps as
Interpretation
Pearson)
there is no relationship between bake_off searches and receipes searches
H0: ...................................................................................................

There is significant relationship between bake_off searches and receipes searches

H1: ...................................................................................................

the risk of incorrectly rejecting h0 p=.000 since the risk is below the .05 thoreshold H0 is rejected .
....................................................................................................
there is hence a significant positive moderate relationship ( rs =.635 ) between bake_off searches
....................................................................................................
....................................................................................................
and receipes searches
....................................................................................................
....................................................................................................
....................................................................................................
....................................................................................................

(Lund & Lund, 2018)11 12

 © Charfi 2021 © Charfi 2021

How to run Spearman Rho on SPSS Let’s recap!

• Analyse  Correlate  Bivariate • Pearson parametric
• Failure to guarantee assumptions (or for
ordinal data)? Choose non parametric statistics

spearmen rho
....................... Eta
....................

kendall tau
..........................
13 14

© Charfi 2021 © Charfi 2021

3. Chi Square χ2 and associated
The logic behind Chi2
measures (nominal/nominal)
• Chi2 is a statistical test that measures the
relationship between two
nominal variables 5 5 10
................................................................
• It builds on contingency tables
(crosstabulation)
5 5 10
• The formula for χ2 is:
10 10 20
(Bryman & Cramer, 2001)
15 16
 © Charfi 2021 © Charfi 2021

Example contingency table output

10 0 10
expected 5

0 10 10

10 10 20
Chi2 compares the entered values with expected values
(expected = if no link exists between the variables)
17 18

© Charfi 2021 © Charfi 2021

How to run Chi2 on SPSS Needed options

19 20
 © Charfi 2021 © Charfi 2021
Interpret significance (Pearson Chi-Square
Interpreting output Asymp-Sig)
• Before interpretation: check for minimum expected count

21 22

© Charfi 2021 © Charfi 2021

Interpretation
• H0: There is no significant link between the use of • If p<.05, it is important to measure the
payment mode and social class.
• H1: There is a significant link between the use of strength of the relationship.
payment mode and social class. • Unlike, Pearson, Spearman, Kendall, etc.
Chi2 does not provide indications on the
• The risk of incorrectly rejecting H0 is almost null
(p=.001); therefore, H0 can be safely rejected. It is strength of the relationship of the tested
concluded that there is a significant link between variables. Additional tests should be run for
the use of payment mode and social class..
this purpose.
23 24
 © Charfi 2021 © Charfi 2021
Complementary tests for relationship
strength
- Contingency Coefficient (value between 0 and 1)
- Phi (value between -1 and 1)
- Cramer’s V (value between 0 and 1) (used with
tables larger than 2x2)
• Interpretation:
– Closer to 0: ............................
weak or no correlation
– Around 0.5: ...........................
moderate correlation
– Closer to 1: ............................
srrong correlation

25 26

© Charfi 2021 © Charfi 2021

Chi2 assumptions Activity

• Sample randomly selected • An American administrator is interested in knowing whether
American school teachers are racist. He would like to know if
• Independence of observation there is a relationship in the mentioned context between a child’s
racial/ethnic background and their likelihood of being suspended
• A large sample from school?
• For contingency tables with df=or>1,
– No more than 20% of cells should have an expected • Open “PracChEthn”
count <5
– Minimum expected count = or > 1 • Perform the relevant statistical analyses to test the relationship
between the “RACEETH” and “SESUOUT”.
merge category , increasing the sample
# Solution:......................................................................... • What can you conclude?
according to the chi square test table two celles (20) have less than expected count 5
27
the maxium limit is 20 percent so the assumption is respected 28

the minimum expected count is 1.4 > 1 so the assumption is respected

h0 there is no relationship between the race of students and number of their exclution
h1 there is a relationship between the race of students and number of their exclution
 © Charfi 2021 © Charfi 2021

The 2nd recap! Activity

• Interpret sig Chi2 • Open “PracChEthn”
– Exception for 2x2 tables: • Is there a link between
• Interpret significance with continuity correction gender “SEX” and being
• If the sample size is small (40 or fewer), interpret expelled from school
significance using Fisher “SESUOUT”?
• If p<.05, interpret relationship strength • Perform the necessary tests
– ........................................
contingency coeff phi
or .............. for 2x2 tables and check test assumptions.
carmer's V
– ........................................ for tables larger than 2x2 • What can you conclude?
29 30

according to carmer's v ( the table is bigger than 2x2) the result is .112 the relationship is weak so the school administration could design flayers

and training sessions to eliminate racism

© Charfi 2021

References
Bryman, A., & Cramer, D. (2001). Quantitative data
analysis with SPSS release 10 for Windows. East
Sussex: Routledge.
Gray, C. D., & Kinnear, P. R. (2011). IBM SPSS 19
statistics made simple. New York: Psychology Press.
Lund, A., & Lund, M. (2018). Spearman's Rank-Order
Correlation using SPSS statistics Retrieved
15.11.2020, from https://statistics.laerd.com/spss-
tutorials/spearmans-rank-order-correlation-using-
spss-statistics.php
31