Lecture 2 Flashcards
(34 cards)
What does the latent variable model assume
that correlations between observable symptoms, like debts, family fall out, are driven and explained by their dependence on a common construct. So for example, if the level of addiction increases, the observable symptoms will also increase as a result
But in what way can you also look at psychological costructs?
In a network model
In a network model what are the bridge symptoms
the overlapping symptoms and the external factors that may contribute to the comorbidity between the two
What did low connectivity showed in the simulation?
linear activation: when something bad happends, it causes presence of symptoms. When this external factor is gone, the presence of symptoms is gone.
The high connectivity settings showed something interesting
when an external factor is present, it causes the presence of symptoms. When this external factor is gone, the symptoms remain!
One of the only models of mental health named something that is the defining attribute of disorder states
Hysteresis
Weakly connected networks are …. and show …
resilient
spontaneous recovery
Strongly connected networks display hysteresis.. what does that mean
the phenomenon where a systems response to changes depends on its history, and reversing a change doesn’t immediately reverse the system’s state
Summarise the ‘mental disorders as aternative stable states’ theory
mental disorders are alternative stable states in a symptom network. Thus, mental disorders are due to hyperconnectivity of the symptom network and perturbations
this leads the network to get stuck in the alternative stable state of mental disorder. Wheter this shift is permanent depends on the size of the hysteresis effecft
What is IsingFit
The first software designed to fit these networks for binary data
A network grows quickly
a network with p nodes features p*(p-1)/2 connections and p thresholds.
How does a network of binary variables define a contingency table?
Of 2”p. So for 15 variabes we have 32769 cells
So what can be seen as a problem?
too many cells. Because of large datasets
What is a solution for the large databases and the fact it would not fit?
- local estimation of smaller parts of the network
- using penalised likelihoof methods like lasso to help with variable selection
- after estimating these small parts, stitch them together to reconstruct the full network
What is the primary aim of Van Borkulo’s eLasso algorithm?
To construct a network model by identifying strong relationships between variables using penalized logistic regression.
In the eLasso algorithm, how are variables used in regression?
Each variable is treated as a dependent variable in a logistic regression, with all others as predictors.
What does it mean for variable X to be in the “neighbourhood” of variable Y?
X is included as a predictor in the penalized regression model for Y.
What kind of regression is used in Van Borkulo’s algorithm?
Logistic regression with Lasso penalization.
When do two variables get connected by an edge in the network?
When each appears in the other’s neighbourhood (mutual prediction).
What is the role of the Lasso penalty in eLasso?
It forces weak connections to zero, leaving only strong, relevant predictors.
What is the final outcome of the eLasso procedure?
A network model showing only strong predictive relationships between variables.
What are “Rorschach networks”?
Network models that might seem meaningful but could reflect noise or overfitting — leading to over-interpretation.
Why is replicability a concern in psychological network models?
Because they involve many parameters and complex statistical processes.
How can we assess whether a network model is robust?
By simulating data from the estimated network and checking replicability across samples.
