Measuring behavior summary
Chapter 1 : Introduction
Behavior -> Actions and reactions of an organism or group. This counts for living organisms, robots,
virtual agents etc.
Direct observation -> The behavior does not need to be directly observed, but that the variables
recorded must relate directly to the actual performance of the behavior in question. However, in
animals the direct observation of the behavior is the only option.
Qualitive = richer impression of the behavior
Quantitative = easier to draw conclusions from the data because they are more transparent and
reproducible
Why observe behavior:
- Major output of the brain
- Gives insight into cognitive processes underlying it
- Major means by which organisms adapt to and affect their physical and biological
environments
What is difficult about observing behavior?
- Temporal component – it unfolds over time
- Behavior is not discrete making it hard to determine when it starts and ends
- Behavior patterns are extremely complicated
- Behavior changes in response to stimuli
- Highly variable
o Between individuals
o Within individuals
Steps involved in studying behavior :
1. Ask a question
2. Hypothesis
3. Predictions
4. Identify and define behavioral matrices
5. Method
6. Designing
7. Ensure legality and ethicality
8. Pilot studies
9. Preregister (otherwise it is exploratory)
10. Collect data
11. Analyze
12. Communicate the results
,Chapter 2 : Science and truth
Goal of science -> scientific theory which are scientific truths
Terms:
- Effect size = Quantifies the size of difference in means of two groups or the strength of a
correlation
o Standard effect size = Expresses this strength in standard deviations
- P-value = probability of obtaining an effect at least as large as the observed effect
considering the effect is 0. Is it a coincidence or
not?
Replication crisis = A lot of behavioral studies have not been
replicated because it is so complicated and now there is a
lot of incorrect and misleading information out there
- Reproducibility = Do the same results and conclusions come out of the same data set
o Can differ due to the researchers degree of freedom (can eliminate data etc.)
- Literal replication = Is the same result obtained if the study is repeated using the same
protocol yielding new data
o In behavior two studies are never really identical
o May give rise to unjustified confidence about the generality of findings that rely om
the methodology
- Constructive replication/triangulation = attempting to produce convergent results that lend
support to the initial study finding but using different methodology.
o This is very strong evidence if they come to the same conclusion since this is
completely independent of any methodological coincidences
Why does replication sometimes fail?
- Sometimes not literal replication giving rise to moderators
o Different reaction norms -> individuals, populations etc. reacting differently from
each other to the same situation
o Standardization fallacy -> Here some things are over standardized and therefore
create different results. E.g. one laboratory always tests mice in the morning the
other in the afternoon, this will make a difference. If they both test them the entire
day their average would be the same
- Low statistical power
o Good scientific methodology aims to minimize false-positive (controlled by p-value)
and false-negatives (controlled by sample size). When either of these is not very
good they have a low statistical power due to constraints on money, time or
availability of subjects.
o Winners curse = if a small study discovers a true-positive the effect size may be
exaggerated. This can be prevented by repeating the study a lot more often with
bigger sample sizes
Positive predictive value (PPV) = Probability that a positive research finding is
a true-positive result. Depends on
Probability of the finding being true before conduction
Statistical power
Threshold for significance
,Low-powered studies + small sample size -> more likely unreliable findings
- The probability of false-negatives is higher
- The probability of a positive effect being true-positive is lower
- Estimates of size of true-positive is overestimated
Publication bias = Tendency for positive results to be published and negative results to remain
unpublished (file drawer problem)
- Results considered to be implausible are less likely to be confirmed by successful replication
o Solution ; adopting different P-value criteria
Pp-hacking = Researchers may (un)consciously exploit flexibilities in data collection, analysis and
reporting in order to get a positive result
- Data collection stops after finding an effect
- Excluding all outliers
- Dependent variables are switched
- Data dredged until significance is found
HARKing = Negative results are suppressed and hypothesis is post-hoc reformulated to fit the data
(Hypothesising after results are known)
Identifying reliable sources of evidence:
1. The evidence pyramid
Hierarchy of evidence that is strongest to weakest. So weight
is given to the research. It can be used to evaluate the quality
of evidence.
2. Types of publication
Where a study is published can also say something about how
reliable the results are.
- Peer-reviewed
- Grey-literature -> research produced by organizations or individuals outside the traditional
academic publishing channels
- Impact factor of the journal (how often cited in other papers)
- Retraction -> findings are no longer trustworthy. Higher this rate the less trustworthy the
information in the journal is
3. Open science movement
- Preregistration -> summary of methods, hypothesis etc. should be provided before data
collection. Problem is there is less creativity.
- Registered reports -> Researcher writes and introduction and method and this will be peer-
reviewed. If it is accepted they can later publish their results regardless of if they are
positive or negative
Overall this movement promotes openness when it comes to how the research is done. That
everyone should be able to look into it.
, Chapter 3 : Choosing a research question
Observing behavior without a research question gives two problems:
- Behavior is complex and it is impossible to measure all aspects with the same accuracy and
precision. You need to know what the put the focus on.
- The more things measured, the higher the risk of getting false-positive results by chance.
Tinbergen’s four questions
1. Mechanism
How does it work? -> Question of proximate causation. These concern the internal and external
causal factors that elicit and control the behavior from moment to moment. Physical and social
stimuli from the environment and the underlying psychological, physiological and neurobiological
mechanisms that regulate an organism’s behavior.
2. Development
How did it develop? -> Question of ontogeny. These are about how the behaviors are assembled and
how it changes during the lifetime of an individual from conception to death. It is about internal and
external factors from the environment shape development.
3. Function
What is it for? -> Question of current utility/adaptive significance. It concerns the current use of
behavior in helping the individual survive and reproduce in its current physical and social
environment. These behaviors may arise from learning and cultural processes.
4. Evolution
How did it evolve -> Question of phylogeny. How the behavior arise during the evolutionary history
of species.
Similar to function but for example : Stopping at red light
Development -> have learned this rule from media/driving instructors
Function -> Who do not stop at the light are able to crash or stopped by police
Evolution -> Red light became the universal sign for traffic to stop
Usually these questions all address a
part of a problem for instance. There is
no right or wrong, but they all address
a different viewpoint.
Different levels of causal explanation
To find out which discipline is the best to answer a certain question, the question needs to be known.
It really depends on the question and the aim of the research which one is best.
Choosing the right subjects
Especially with animal species this can be difficult. Questions that can be asked to determine which
species is the best option are:
- Does an extensive biological and behavioral literature exist for this species?
- It is natural behavior well suited to the particular problem to be investigated?
- What are its life-history characteristics, such as gestation period, age of independence and
age of sexual maturity?
Chapter 1 : Introduction
Behavior -> Actions and reactions of an organism or group. This counts for living organisms, robots,
virtual agents etc.
Direct observation -> The behavior does not need to be directly observed, but that the variables
recorded must relate directly to the actual performance of the behavior in question. However, in
animals the direct observation of the behavior is the only option.
Qualitive = richer impression of the behavior
Quantitative = easier to draw conclusions from the data because they are more transparent and
reproducible
Why observe behavior:
- Major output of the brain
- Gives insight into cognitive processes underlying it
- Major means by which organisms adapt to and affect their physical and biological
environments
What is difficult about observing behavior?
- Temporal component – it unfolds over time
- Behavior is not discrete making it hard to determine when it starts and ends
- Behavior patterns are extremely complicated
- Behavior changes in response to stimuli
- Highly variable
o Between individuals
o Within individuals
Steps involved in studying behavior :
1. Ask a question
2. Hypothesis
3. Predictions
4. Identify and define behavioral matrices
5. Method
6. Designing
7. Ensure legality and ethicality
8. Pilot studies
9. Preregister (otherwise it is exploratory)
10. Collect data
11. Analyze
12. Communicate the results
,Chapter 2 : Science and truth
Goal of science -> scientific theory which are scientific truths
Terms:
- Effect size = Quantifies the size of difference in means of two groups or the strength of a
correlation
o Standard effect size = Expresses this strength in standard deviations
- P-value = probability of obtaining an effect at least as large as the observed effect
considering the effect is 0. Is it a coincidence or
not?
Replication crisis = A lot of behavioral studies have not been
replicated because it is so complicated and now there is a
lot of incorrect and misleading information out there
- Reproducibility = Do the same results and conclusions come out of the same data set
o Can differ due to the researchers degree of freedom (can eliminate data etc.)
- Literal replication = Is the same result obtained if the study is repeated using the same
protocol yielding new data
o In behavior two studies are never really identical
o May give rise to unjustified confidence about the generality of findings that rely om
the methodology
- Constructive replication/triangulation = attempting to produce convergent results that lend
support to the initial study finding but using different methodology.
o This is very strong evidence if they come to the same conclusion since this is
completely independent of any methodological coincidences
Why does replication sometimes fail?
- Sometimes not literal replication giving rise to moderators
o Different reaction norms -> individuals, populations etc. reacting differently from
each other to the same situation
o Standardization fallacy -> Here some things are over standardized and therefore
create different results. E.g. one laboratory always tests mice in the morning the
other in the afternoon, this will make a difference. If they both test them the entire
day their average would be the same
- Low statistical power
o Good scientific methodology aims to minimize false-positive (controlled by p-value)
and false-negatives (controlled by sample size). When either of these is not very
good they have a low statistical power due to constraints on money, time or
availability of subjects.
o Winners curse = if a small study discovers a true-positive the effect size may be
exaggerated. This can be prevented by repeating the study a lot more often with
bigger sample sizes
Positive predictive value (PPV) = Probability that a positive research finding is
a true-positive result. Depends on
Probability of the finding being true before conduction
Statistical power
Threshold for significance
,Low-powered studies + small sample size -> more likely unreliable findings
- The probability of false-negatives is higher
- The probability of a positive effect being true-positive is lower
- Estimates of size of true-positive is overestimated
Publication bias = Tendency for positive results to be published and negative results to remain
unpublished (file drawer problem)
- Results considered to be implausible are less likely to be confirmed by successful replication
o Solution ; adopting different P-value criteria
Pp-hacking = Researchers may (un)consciously exploit flexibilities in data collection, analysis and
reporting in order to get a positive result
- Data collection stops after finding an effect
- Excluding all outliers
- Dependent variables are switched
- Data dredged until significance is found
HARKing = Negative results are suppressed and hypothesis is post-hoc reformulated to fit the data
(Hypothesising after results are known)
Identifying reliable sources of evidence:
1. The evidence pyramid
Hierarchy of evidence that is strongest to weakest. So weight
is given to the research. It can be used to evaluate the quality
of evidence.
2. Types of publication
Where a study is published can also say something about how
reliable the results are.
- Peer-reviewed
- Grey-literature -> research produced by organizations or individuals outside the traditional
academic publishing channels
- Impact factor of the journal (how often cited in other papers)
- Retraction -> findings are no longer trustworthy. Higher this rate the less trustworthy the
information in the journal is
3. Open science movement
- Preregistration -> summary of methods, hypothesis etc. should be provided before data
collection. Problem is there is less creativity.
- Registered reports -> Researcher writes and introduction and method and this will be peer-
reviewed. If it is accepted they can later publish their results regardless of if they are
positive or negative
Overall this movement promotes openness when it comes to how the research is done. That
everyone should be able to look into it.
, Chapter 3 : Choosing a research question
Observing behavior without a research question gives two problems:
- Behavior is complex and it is impossible to measure all aspects with the same accuracy and
precision. You need to know what the put the focus on.
- The more things measured, the higher the risk of getting false-positive results by chance.
Tinbergen’s four questions
1. Mechanism
How does it work? -> Question of proximate causation. These concern the internal and external
causal factors that elicit and control the behavior from moment to moment. Physical and social
stimuli from the environment and the underlying psychological, physiological and neurobiological
mechanisms that regulate an organism’s behavior.
2. Development
How did it develop? -> Question of ontogeny. These are about how the behaviors are assembled and
how it changes during the lifetime of an individual from conception to death. It is about internal and
external factors from the environment shape development.
3. Function
What is it for? -> Question of current utility/adaptive significance. It concerns the current use of
behavior in helping the individual survive and reproduce in its current physical and social
environment. These behaviors may arise from learning and cultural processes.
4. Evolution
How did it evolve -> Question of phylogeny. How the behavior arise during the evolutionary history
of species.
Similar to function but for example : Stopping at red light
Development -> have learned this rule from media/driving instructors
Function -> Who do not stop at the light are able to crash or stopped by police
Evolution -> Red light became the universal sign for traffic to stop
Usually these questions all address a
part of a problem for instance. There is
no right or wrong, but they all address
a different viewpoint.
Different levels of causal explanation
To find out which discipline is the best to answer a certain question, the question needs to be known.
It really depends on the question and the aim of the research which one is best.
Choosing the right subjects
Especially with animal species this can be difficult. Questions that can be asked to determine which
species is the best option are:
- Does an extensive biological and behavioral literature exist for this species?
- It is natural behavior well suited to the particular problem to be investigated?
- What are its life-history characteristics, such as gestation period, age of independence and
age of sexual maturity?
