NetworksVizInference
A repository for all the things relating to visual inference and model diagnostics on network models.
Outline
- Motivation and Visual inference Intro
- Define "visual" effects
- Model Diagnostics - what we want to do
- Set up and describe online experiments
- Discuss and analyze experiment results
- Draw conclusions
Other notes:
- need a lot of data
- need a lot of significant models
- currently working on an R package to pull this all together. This might mean another repo -- sorry.
"Chapter 2"
- Start with senate data (probably need to remove HRC): higher threshold than .25 for WPC to make network smaller
- Find the significant effects of the SAOM eval effects
- Fit various models many times - simple, one interaction, one interaction + one other, two interaction (party, bills), two interaction + one other, etc, ones I think make sense
- Get mean parameters from each of those models to use in simulation in the future.
- Simulate "data" from a bunch of different models: a. The models we care about from step 3 b. Same models from (a) but with various changes to the parameters. Like What? i. make rate parameters constant over all periods ii. double / halve / increase by 50% / etc various parameters one-at-a-time. iii. Switch signs of parameters
- Do I need to refit the model to the data in between getting this "data" and simulating for the lineup? (I think maybe I do)
- Once I have 100 (?) data sets from each model in 4, create a lineup of each of those data sets among even more sims (M = 12) for use in an mturk study
- "Easy Lineups" Idea: put the data from the one model among a bunch of sims from "opposite" models (e.g. parameter value switched signs, doubled, etc.) or models with totally different parameters
- Novel Lineup Idea (?): store data from all waves, view 6-10 rows of dynamic networks, all waves.