hannesdatta / covid-19-book-consumption Goto Github PK

Please test-run the workflow, report any occurring issues, and come up with a list of suggestions on how to improve the workflow.

Minimum requirements are:

• download raw data
• reproduce all results
• consistently use make
• use versioning

• Note any selection rules applied on users
• Collect some descriptive on the user level

@mpachali

provide by @mpachali to @srosh2000

We will work on this project largely in Overleaf for writing. Please setup the Overleaf paper and make it a living document in which we can view the most recent figures and tables. Eventually, the paper will evolve there.

• Data on published books and their publishers
• Rankings in terms of how famous these publishing houses are
• Book genre
• Status of the author i.e. experience in publishing books [ potential metric?]

• Covid impact on consumer side: During Covid, bookstores in Germany and Belgium remained open while they were closed in the Netherlands.
  • This led to higher book discovery in Germany and Belgium relative to NL
  • Consumers in NL stuck to previous preferences
• Covid impact on supply side of book market: some publishers stopped making new deals i.e. accepting new manuscripts. This has an impact on book market quality and variety.
  • What would be some supply side metrics? Do we have access to them?

We have created a project board to prioritize tasks, but I haven't been able to link it to this repository yet. Can you somehow establish the connection, please?

Please evaluate the following model ideas, and develop a plan for models (e.g., a combination of what we list below) need to be estimated as a next step.

1.) Have you thought about a Seemingly Unrelated Regression (SUR), as errors are very likely correlated and estimated regressions separately is a hassle
2.) I don’t like so much that covariates (like gender, country, etc.) are interactions. It may also be worthwhile estimating an individual model (Hierarchical Linear) with covariates included to leverage the richness of the individual-level data
3.) Shouldn’t we be able to estimate a Diff&Diff as in Sim et al. to nail the Covid effects? Not sure whether it is useful though…
4.) Also, we need to consider different ways of estimating long-term effects. An error correction model (see Hannes' 2022 JMR) may be one straightforward way to do this. Has this model been also estimated at the consumer-level before (probably by Harald van Heerde at some point in time?). Are there any alternative/good ways to model long-term effects?

Long-term effects are important because consumers might dynamically optimize, i.e. increase consumption during lockdown and reduce after to seek variety. You also see this over the seasons (Figure 3+4). However, the net effect is still positive (Figure 4), suggesting that Covid engaged more people to continue reading after lockdowns. ****

Deliverable:

• Draft an analysis framework (i.e., not run these models, but suggest very concretely which models you WOULD run).

@mpachali

The cross-country variation in the data is quite rich: can you please generate a list of potential covariates available at the country level that could explain either (a) the size of the Covid-19 effects ("increase in book consumption moderated by..."), or the (b) duration of (long-term) effects (e.g., "effects last longer because consumers in country A have this and that attribute).

A few ideas:

• Hofstede
• European Social Survey (I think that's what it is called)

Also check out Jan Benedict Steenkamp's work on international marketing. Inge Geyskens also uses a lot of country characteristics.

The raw data has been documented (see PDF here on the repos). Please (a) check the documentation (do we understand everything), and (b) come up with a list of questions to ask to Mike about how the data was generated.

The deliverable is an updated documentation, to be committed to this repository (let's have it as a markdown document).