Lecture 2 Flashcards
The first part of lecture 2 is about the network approach, why is this approach important in theory construction?
Lecturer shows an example of a latent variable model, and explains that this is too simple of an approach, the network models show a more complete and complicated model of actual symptoms and causes
By making a latent variable model a network model, what happens to indicators of psychological constructs?
They become freestanding causal entities in the system
What is a latent variable model?
One disorder causes all of the symptoms
To which model do psychologist bias toward?
Latent variable model
What are four key factors in a symptom network?
Symptom network A, bridge symptoms, symptom network B and external factors
What are external factors in symptom networks?
They are the “shocks” of misery that a person gets in their life
What are bridge symptoms in symptom networks?
They are the ones that connect different symptom networks (A and B, etc.)
What happens (generally) in symptom networks when one symptom rises in severity?
The probability that a connected system also rises a bit increases
What is the difference between low and high connectivity in symptom networks? (do NOT talk about state differences)
Low connectivity can withstand many shocks, aka, external factors do not affect them as much
How do low and high connectivity differ in their patterns (again, NOT states)?
Low connectivity shows a linear pattern (bell curve in simulation). Big shocks = big reaction and small shocks = small reaction. When external field is normal again, so will the reaction.
High connectivity it can sustain its own activity, meaning that even if the external field is normal again, reaction will not necessarily be
What is connection strength (in network model)?
Degree to which it is the case that one symptom triggers another
There is an explanation taken from a different field to explain what happens with the pattern in high connectivity individuals, what is this?
Hysteresis; a physics concept where, for instance, something becomes a liquid at 70 degrees, but only turns back to a solid at 30 degrees.
There are “states” one can be in in these symptom models, explain this and the difference between high and low connectivity
Low connectivity has a baseline state, it goes out of this baseline when external activation is great enough, however, when this is normal again, they will also return to their baseline state.
High connectivity, when external activation is great enough, falls down into a different state, and it takes much more effort to get back to the baseline state
(figures may make this clearer)
What is the theory of the symptom network?
Mental disorders are alternative stable states in a symptom network and are thus due to local hyperconnectivity of the symptom network and external factors. As such, one can get stuck in a disordered stable state
Using networks has its downsides, what is the main problem?
They grow at insane rates
There is a solution to the problem that network models poses, what is this?
Local estimation of network neighborhoods
through a penalized likelihood (lasso), and stitching the whole network back together
(also see figure for clarity)
Explain simply how the Elasso algorithm works
One by one, each variable is a dependent variable in a logistic regression, with all the other variables as independent
gives regression equation, optimized using a penalization function (lasso)
When a variable X (independent) is included in the prediction function for Y (dependent), then we say that X is in the neighbourhood of Y
When they are estimated to be in each other neighbourhoods, they are connected
What do red lines in model networks indicate?
negative correlations
One of the examples of these network models currently in practice is one with a treatment node, explain how the treatment is connected to the symptoms
Before the first trials it is not connected at all (randomized treatment trials), over time the symptoms become negatively correlated with treatment
What are criticisms against the network model? (2)
- Networks are susceptible to overinterpretation because you always see a pathway (it is likened to the rosach exercise, where everyone sees some type of shape)
- Inability to replicate
Lecturer had response to one of the criticisms against network models, which is?
Replicability, they say it is possible to assess how well a network does so by using simulations to assess robustness
The lecture uses ecological systems as an example as to how network models may be used in practice, what do they take from ecology?
In ecological systems, when it is on the bring of collapse (transition point) it fires out signals that can be picked up before the collapse
What is the term for the concept taken from ecology?
Critical slowing down; if a system is near a transition it loses resilience and takes longer to recover from shocks (autoregression?)
What happens when a system is in their transition?
It becomes more predictable
