Measuring behavior and hypothesis testing Flashcards
Planning an experiment
Research question
Hypothesis driven research
Predictions
Descriptive projects
Provide information about a topic through summaries and descriptions of the main features of a dataset
Inferential projects
Test hypotheses and predictions, and make inferences and generalizations about larger population based on data collected from a sample of that population –> hypothesis driven
Scientific method (inferential projects)
Observation/question –> Research topic area –> Hypothesis –> Test with experiment –> Analyze data –> Report conclusions –> Repeat process
You observe phenomenon and ask yourself questions
Make hypothesis from this question = supposition to explain observation
Formulate a prediction = logical consequence of the hypothesis before the outcome of the experiment is known
You build an experiment to test the
hypothesis according to your prediction = you collect quantitative data that can be compared in a controlled way
You analyze your data based on:
- several factors and levels
- control group(s)
- a statistical approach
You conclude, which may lead
to a new question ! It is never ending, that’s the
beauty of it !
Does hypothesis testing require an experimental approach?
No, but it always requires a quantitative approach
Field work or comparative analysis of literature data can also be hypothesis-testing and generate important insight
Hypothesis
Supposition or proposed explanation made on the basis of limited evidence as a starting point for further investigation.
–> ‘we hypothesize, based on the literature, that ‘chaos’ in a sound serves as a repellent signal’
Prediction
Logical consequence of the hypothesis before the outcome of the experiment is known.
–> ‘If our hypothesis is true, we predict that sheep will move away from the speaker more when more chaotic sounds are being played compared to less chaotic sounds’
Building an experiment
Changing one factor at a time to avoid confounding effect
Create a control group to compare with
Have a sufficient amount of sample to have reliable conclusions
Report all the procedure to allow replication of your protocol
Need to be feasible
- We need to know how to quantify the response!
The null hypothesis H0/ The alternative hypothesis H1
- H0: Experimental group is not different from control group
- H1: Experimental group is different from control group
Sample size matters: The statistics
Accepting a hypothesis needs a significant P-value in data analysis P < 0.05 generally
−
But P > 0.05 does not imply the null hypothesis is correct, just that we cannot reject it
To accept a null hypothesis you need to know the power of the test
That may require more samples = more work
1) Sampling rules for studying behavior
1) Ad-libitum sampling –> All (conspicuous behaviors)
2) Scan sampling –> All (conspicuous) behaviors, groups of individuals at specific time intervals
3) Behavior sampling –> Specific behavior
4) Focal sampling –> Specific individual
Ad-libitum sampling - what and who?
What: All behaviors (that a visible)
Who: Few/all individuals in the group (that are visible)
–> preliminary observations
–> conspicuous behaviors
Eating, resting, self-grooming
Scan sampling - what and who?
What: All behaviors (that are visible)
Who: Few/all individuals in the group at specific time intervals
–> Conspicuous behavior
Time point 1:
- Eating (15 cows)
- Resting (11 cows)
- Self-grooming (3 cows)
Behavior Sampling - what and who?
What: One specific behavior
Who: Few/all individuals in the group
–> limited number of behaviors
Self-grooming: time point 1
- 3
Focal sampling - what and who?
What: All behaviors
Who: One specific (focal) individual
–> limited number of individuals
Time point 1:
- Horse 2: aggresses horse 3
2) Recording rules for studying behavior
1) Continuous recording –> continuous period
2) Time sampling –> instantaneous/one-zero
Continuous recording - when?
Record continuously during a pre-defined period
–> time consuming, but all instances of behaviors scored, real durations and frequencies
Time sampling - when?
Record periodically at regular intervals
–> Less time consuming, but instances of behaviors missed and not real durations/frequencies
3) Types of behaviors
Events: calling, kicking, biting, head shaking (short duration) (< 1s) –> rate of occurrence (nb/s)
States: walking, lying, eating (long duration) (> 1s) –> % of time (%)
Latencies –> time (s)
Frequency (rate of occurrence)
Number of times behaviors occurs during the sampling period –> rate (frequency/time) of EVENTS
Duration (time unit or % of sampling duration)
Length of time of a single behavior –> % of time spent (duration/time) of STATES
Latency (time unit)
Time from beginning of recording session/experiment to onset of first occurrence of behavior
4) Defining an ethogram
List and definitions of behaviors to score
–> All observable behaviors performed in the studied situation
–> Clearly defined
–> Can be differentiated
–> Definition ≠ Interpretation
5) Validation
Observations from videos
- High intra and inter-observer reliability
- Observer is blind to the treatment (when possible) –> codes
- Methods are described with enough details
Intra/interobserver reliability
*Intra = within-observer agreement
→ the same person codes the same videos twice
- Inter = between-observers agreement
→ two persons code the same videos
→ correlation (e.g. Spearman, Pearson, Kendall) or ICC
Assessment (R^2)
* Less than 0.4: poor
* 0.4 – 0.59: fair
* 0.60 – 0.74: good
* 0.75 – 1.00: excellent
