Download as pdf or txt
Download as pdf or txt
You are on page 1of 30

Outils de conceptualisation

Cours 4

MASTER SCIENCES, TECHNOLOGIES, SANTE 2023-2024

Mention Ergonomie

Louise Kirsch
Integrative Neuroscience and Cognition Center
louise.kirsch@u-paris.fr
Group work

What we have seen – what you should have in your report:


RESEARCH QUESTION & HYPOTHESIS OUTLINE
A research domain
A research question
An hypothesis on the research question
EXPERIMENTAL PLAN OUTLINE
An experimental sample (human or non-human)
An experimental plan: how to measure your DV and IV; which kind of study design to use

+ check for possible confounds (cf. previous lecture).


Group work
• Let’s share your research question and hypothesis
• 15 min to gather the info by groups and put on wooclap
• 2 min presentation + 3 min discussion per group
Part IV - Data Analysis
Data analysis
Data analysis

…What is ‘data’ ?

The variables that you have measured (i.e. your measurements)


Information about your variables
PERFORMANCE
RATING
INSTRUMENTAL
SURVEY ANSWERS MEASUREMENTS Ex.: performance
Ex.: Low back pain rating in the
TEST RESULTS Ex.: Measurement simulation-training
survey of movements
Ex.: Episodic study
memory test during sitting
position through
sensors
Data analysis
In quantitative research, data take the form of a table
You associate your participants to the measures obtained from them

Ex. of data from Outils_1


Examples of possible
measurements
Data analysis: exercise SURVEY ANSWERS
Ex.: Low back pain
1. Exercise : which measurements would you do in your Experiment ? survey

- Brainstorm PERFORMANCE
- Write an outline in your document RATING
Ex.: performance
⇢ When writing a research plan, it is necessary to know precisely which kind of data you rating in the
will obtain with your experiment simulation-training
study
NB This is known prior to conducting the experiment - The only thing that we do not
know are the results INSTRUMENTAL
MEASUREMENTS
2. Exercise : which data would you have if you run your Experiment ? Ex.: Measurement
of movements
NB Your data depend on your measurements. Look at the measurements that you chose and during sitting
note down what kind of data they produce. position through
DATA FROM A SURVEY OR TEST = answers (values) that may be: numeric/not numeric; sensors
hierarchical (e.g. intensity of low back pain from 1 to 5) or non-hierarchical (e.g. voting
intentions) TEST RESULTS
INSTRUMENTAL MEASUREMENTS = depends on the equipment used Ex.: Episodic
etc. etc. memory test
Data analysis
Standard deviation
Data ⇢ Data analysis (= statistical analysis) ⇢ Results !

DESCRIPTIVE STATISTICS

Mean
Median
Dispersion
Data analysis
INFERENTIAL STATISTICS

YOUR RESEARCH HYPOTHESIS = ‘ALTERNATIVE HYPOTHESIS’


E.g. There is a relationship between static sitting habits and low-back pain: persons with static sitting
habits have higher low back pain
Alternative to what?
Null Hypothesis: There is no significant relationship between the variables given (H0)
Alternative Hypothesis: There is a significant relationship between the variables given (H1)
Null hypothesis Alternative hypothesis
Hypothesis that the researcher tries to Hypothesis that the researcher tries to
disprove prove

• Why is it called ‘Alternative’? The hypothesis has yet to be proved with sufficient pieces of evidence
• The study will tell you if you can reject the null hypothesis
Data analysis
Ex. “Low back pain study”
Null hypothesis Alternative hypothesis
Hypothesis that the researcher tries to Hypothesis that the researcher tries to
disprove prove
There is no relationship between sitting Sitting habits do influence low back pain
habits and low back pain in call-centre in call-centre employees
employees
Data analysis

When can you reject the null hypothesis?


⇢ When you find statistically significant relationships between your variables

What is a statistically significant result?


⇢ A result that can not be explained by chance alone (that did not happen by
chance)

How do you know if your results happened by chance or not?


⇢ Statistic models
Data analysis

Statistic test tell if the results (1) happened by chance = not significant or (2) happened because your
hypotheses on the relation between variables are true = significant

STATISTICS (IN HUMAN SCIENCES)


Since we cannot study variables in their global distribution, we build a
statistical model where we test these variables.
A mathematical model of the section of the world we are interested in.

How do statistical tests work? Stat. tests calculate:


• A test statistic = a value describing how much the relationship between your variables differs
from the null hypothesis (how much the relationship static sitting/low back pain differs from the
hypothesis that there is no relation)
• A probability value (p-value) = a value describing how likely it is to observe your results by
chance alone (if the null hypothesis was true). The lower the better (how likely it is to observe the
relationship static sitting/low back pain, that is in the data, by chance alone)
Data analysis
Data ⇢ Data analysis (= statistical analysis) ⇢ Results
How to choose a stat. test? Depending on the relationship between the variables you want to test
SOME COMMON STATISTICAL TESTS IN HUMAN SCIENCE
Difference between 2 groups T-TEST Are ice-cream sold more in summer or in winter? T-test (ice-
cream sold in summer vs ice-cream sold in winter)
Difference between > 2 groups; ANOVAs Are ice-cream sold more in spring, summer or winter? ANOVA
Predictive value of an IV (or more) on (ice-cream sold in spring; summer; winter). Categorical IV
a DV
Predictive value of an IV/set of IV on (SIMPLE/MULTIPLE) Do external temperature (and country) predict the quantity of
a DV REGRESSION ice-cream sold? SIMPLE REGRESSION (temperature → ice-cream
sold) or MULTIPLE REGRESSION (temperature; gr sugar x ice-
cream → ice-cream sold). Continuous DV
Association between variables CORRELATION Is the quantity of ice-cream sold associated with external
temperature? CORRELATION (quantity of ice-cream sold;
external temperature)

Note: IV= Independent variable (=group/conditions) and DV= dependent variable (what we measure)
Data analysis
DIFFERENCE BETWEEN 2 GROUPS

T-Test (independent samples = different participants in the 2 groups)


Hypothesis: employees in call-centres have higher back pain than employees in a ice-cream shop
DV = back pain; IV = working place (call-centre vs ice-cream shop)
Measurement of back pain level = intensity survey (1-10)

Results:
• test-statistic (t: how far the difference between your means is from the null hypothesis)
• probability value (p: how probable it is to observe the difference between your means if the null
hypothesis is true.
Significance of p is normally set to <0.05

MEAN MEAN

T-Test (paired samples = same participants in the 2 groups)


Ex.: your hypothesis = the same employees will have higher back pain after work than before work
Data analysis
T-test visualization (“Low back pain” study)
Data analysis
ANOVAs (large family)
AIMS : DIFFERENCES > 2 GROUPES; PREDICTIVE VALUE OF Independent Variables (IV - typically categorical)
One-way ANOVA = 1 IV
E.g. Hypothesis = when season is hot, the quantity of ice-cream sold is higher

Two-way ANOVA = 2 IV
E.g. Hypothesis = (hot) season and (southern) geographic location influence the quantity of ice-cream sold

Repeated measures ANOVA = the same group of participants is measured on the same variable several times
E.g. You measure cholesterol levels in a sample at points 0, 2 and 4 months after having changed their diet
Data analysis
ANOVAs (large family)

Results:
• test-statistic F: variance between sample means / variance within the samples
Variance = how far your values are scattered from the mean

• probability value p: how probable it is to observe the difference between your means if the null hypothesis is true
If one or more IV are significant : significant effect of the variable (‘main effect’)
An interesting fact when having > 1 variable : interactions
The effect of the IV on the DV can depend on another IV
E.g. 2 IV = season (hot; cold) and country (South Europe; North Europe); DV = ice-creams sold
You run a Two-way ANOVA
Result: Season has an effect on Ice-cream sold only when Country = South Europe
⇢ There is a significant interaction between Season and Country
Data analysis
ANOVA visualization :
Time needed for three poisons to give effects
DV = time
IV = poison (red; green; blue)

Do we observe significant effects?


Data analysis
REGRESSION
AIMS : PREDICTIVE VALUE OF IV (typically continuous IV)

SIMPLE REGRESSION = 1 IV
E.g. Hypothesis = when temperature is higher, the quantity of ice-cream sold is higher
test-statistic t (as in t-test); probability value p
r2 proportion of the variance for a dependent variable that is explained by an independent variable

MULTIPLE REGRESSION = 2+ IV
E.g. Hypothesis = temperature, amount of sugar in the ice-cream, price of the ice-cream determine the quantity of ice-cream
sold
Data analysis
REGRESSION visualization :

MULTIPLE REGRESSION :
Data analysis
CORRELATION
AIMS : ASSOCIATION BETWEEN VARIABLES (correlation is not causation)

Ex.: in a group of people, those who buy more ice creams also are more frequently affected by sunburn. à Quantity of ice
creams and quantity of sunburns are positively correlated in the group.

Wrong conclusion: eating ice creams causes sunburn (cf. the “crazy correlations” in Outils_3)
Right conclusion: eating ice creams is (for undetermined reasons) associated to sunburn

NB Sometimes, a correlation might indeed signal a causal relation (e.g. we can imagine that the n. of hours spent on the beach
is positively correlated with sunburn) BUT you cannot show this with correlation

• Correlation tells us if two variables tend to change jointly (association) Correlation does not tell what triggers this
association
• There are no dependent / independent variables in correlation, just associated and non-associated variables
Data analysis
CORRELATION visualization :
Data analysis: exercise
Read: “Simulation-Based Training” study (Statistical analysis; Results - help yourself with the Data
analysis section if needed)

Ref: Ontrup et al., (2020) Does simulation-based training in medical education need additional
stressors? An experimental study

Identify in the text

1. The statistical analyses (what type of tests were conducted)

2. The significant/non-significant results

3. Describe the results presented on the figures


Ontrup et al., (2020)
Ontrup et al., (2020)
Data analysis : Other types of results - Low back pain study
Data analysis

Data from our examples papers

Pre-data collection to assign participants to groups:


A = no pain (control group)
B = pain (experimental group) Dependent Variable
Chronic Pain Grade questionnaire Sitting position instrumental measurement
German brief pain inventory 4 indexes of pain (pain questionnaire)
Independent Variable
Sitting position instrumental measurement
4 indexes à composite score
Data analysis: exercise
Exercise : which stat would you propose for your Experiment ?

⇢ When writing a research plan, it is necessary to know precisely which kind of


data you will obtain with your experiment

NB This is known prior to conducting the experiment


The only thing that we do not know are the results
Next steps

• Expected Outcomes (final section of a research proposal)

• How to write your research proposal

⇢ Take all the Outlines we have worked on and put the elements together, as
divided in proper sections (cf. Outils_3)

You might also like