Documenting test queries in web of science

[lindseysschwartz]

"Amazon Flex" + driver

1 result: https://www.webofscience.com/wos/woscc/summary/b8173632-ce3c-4f57-b2ea-6d04c24db605-80bff1a3/relevance/1

[lindseysschwartz]

Uber + driver ((super broad but I was curious))

386 results: https://www.webofscience.com/wos/woscc/summary/58235fbc-9a27-4cd1-91ce-98e124625a3d-80c049f4/relevance/1

• familar friends Hall & Krueger, Roseblat & Stark, Chen (the Uber economist), Katie Wells
• studies on the mental health of drivers..."technostress"; the taxi industry
• case studies in specific countries: Poland, Taiwan, Australia, Brazil, India, Canada -- this seems like a good sign
• case study on the NYC taxi industry conducted by researchers in Milan

[lindseysschwartz]

DoorDash + restaurant

2 results: https://www.webofscience.com/wos/woscc/summary/39b3f3d7-6091-4eab-9150-1f6918436be0-80c10ada/relevance/1

• I expected a small number of results - there may be a better approach for collecting publications that aren't about TNCs

[nniiicc]

I like the idea of going back and forth with terms added... But it may be better to do a query by platform (e.g. delivery vs ridehail)

"Rideshare" OR "Ridehail" OR "Uber" OR "Lyft" OR "Transportation Network Company" - returns 2590 results

"Delivery Network Company" OR "Doordash" OR "Instacart" OR "Uber Eats" - Returns 26 results

"Amazon Flex"- 1 Results

I would add additional companies to each platform type

The other thing to be aware of ...

• A Web of Science search that is limited by the document type "articles" will not pull in white papers, non-peer reviewed pre-prints etc. So if we go to Scholar and search for "Amazon Flex" - we get more results but they look much less relevant.
• Scholar indexes the full text, where WoS often just does the abstract. In some ways WoS is better because it doesn't surface every brief mention, but only focuses on the mention of a keyword or phrase in the abstract (which likely means it is important to the overall study)

[samMint]

WoS search

Query [("rideshare" OR "transportation networking company" OR "Lyft" OR "Uber" OR "Wings") AND ("wages" OR "infrastructure" OR "environment")]

2014–2023 Abstracts

• 531 results
• Including a claim-type facet limits result yield in comparison to just searching by companies. Maybe we want to avoid a claim facet entirely to get the most recall. (I also wonder if I should've expanded the set of claims I included).
• "Environment" captures "built environment" as well.
• Quickly finding examples of sketchy data provenance. TNC data is used, but the researcher doesn't say how they acquired it. (What does ride-hailing mean for parking? Associations between on-street parking occupancy and ride-hail trips in Seattle https://www.sciencedirect.com/science/article/abs/pii/S2213624X21000493?via%3Dihub).
• Some are less vague about pulling TNC data from an API, but it's not clear whether the data is already published in another study or if the researchers have done their own data scraping. (_On the determinants of Uber accessibility and its spatial distribution: Evidence from Uber in Philadelphia_https://wires.onlinelibrary.wiley.com/doi/10.1002/widm.1362). Here, TNC data was used in combination with census data.

I'm going to try some other queries but it's good to get a sense for what's out there.

Quick observation: I tried including "API" in the above search parameters and it cut me down to 2 results (1 relevant -- Equity of access to Uber's wheelchair accessible service https://www.sciencedirect.com/science/article/pii/S0198971521000958?via%3Dihub). I was hoping this would work but nope! I do wonder if including or excluding certain sampling methods could help in the search.

4/12 Update

• attempted using ["Data" OR "API"] in combination with other query facets. Mixed results in terms of yield. Could be a matter of how I was constructing the rest of the search.

• realized a typo in my original query (Wingz not Wings). The corrected search only yielded 128 results. Dropping all claim facets boosted results to 2,553. Including the ["Data" OR "API"] facet collapsed results to 595 from the original 2,553.

• in sum, wondering if it could be worth dropping the idea of a claim facet and exploring other possibilities to make the results more precise.

[lindseysschwartz]

"delivery network company" OR "third-party delivery" OR "DoorDash" OR "Gojek" OR "Shipt" OR "Caviar" OR "Roadie" OR "Jumia Food" OR "Deliveroo" OR "Foodora" OR "Swiggy" OR "PedidosYa" OR "Rappi"

• 1286 results
• Caviar returns studies about caviar the food (273 results are in Marine Biology, and more scattered across Dairy & Animal Sciences, genome sciences, molecular biology, and other physical sciences)
• Eliminating Caviar as a search term yields 85 results - a mix of studies about delivery logistics, economic sustainability of the delivery sector, labor protests - which seems good for not using our "claims" terms
• I'm finding that testing queries is helpful for thinking of exclusion / inclusion criteria
  • the terms Shipt and Rappi returned a few non-relevant studies ("shipt" is an antiquated spelling of shipped; "rappi" is a snake species)
  • Gojek returned a study about the platform, but a Twitter protest analysis (https://www.webofscience.com/wos/woscc/full-record/WOS:000675143500001)
  • This study is qualitative but the researchers interview Deliveroo representatives and shared the manuscript before publishing -- "Contribution to the Sustainability Challenges of the Food-Delivery Sector: Finding from the Deliveroo Italy Case Study" https://www.webofscience.com/wos/woscc/full-record/WOS:000570350800001 -- how do we want to take company reps as a data source into account? To me, this would still fall within "company-sourced data" - researchers are obtaining data directly from companies and have an arrangement about reporting results

[nniiicc]

Next steps:

Decide on a set of queries that includes

• Modifier (e.g. ridehail)
• Add companies or platforms from LS list of most prevalent
• Keep DNC vs TNC datasets separately