Data literacy session plan: reading a real transactional dataset

A librarian-led workshop that uses ALT/FNDATA as a live example for teaching students how to evaluate a dataset before trusting its conclusions. This session is designed to be delivered by library instruction staff, independent of any specific course, and it complements rather than duplicates the course's own data-literacy content (Module 5 in 02_Course_Modules, where it exists). The course module teaches the dataset in service of a finance or economics curriculum; this session teaches the dataset in service of general data literacy, and is suitable for a one-off library workshop, an orientation session, or a drop-in series.

Audience

Undergraduate or graduate students with no prior coding or database experience. Suitable for a general library workshop series, a first-year research methods session, or an add-on to an existing course by instructor invitation. No SQL background is assumed; a short primer is built into the agenda.

Learning objectives

By the end of this session, participants will be able to:

  1. Locate and use a coverage browser to check what a dataset does and does not include before querying it.
  2. Write and modify a simple SQL query in a no-code sandbox environment to answer a specific question.
  3. Distinguish between a stable, well-supported figure and a figure that may be an artifact of incomplete or uneven data coverage.
  4. Explain why the most recent period in a continuously updated dataset can be misleading if read the same way as a historical period.
  5. Identify at least two questions to ask of any dataset before relying on it for a research claim.

Format and length

60 to 90 minutes, in person or synchronous virtual, in a room or platform with individual student devices (a computer lab, laptop cart, or bring-your-own-device session). No installation required; the sandbox runs in a browser.

Timed agenda (90-minute version; trim as marked for 60 minutes)

0 to 10 min: Opening and framing. Ask participants how they currently decide whether to trust a dataset or statistic they find online. Introduce the session's core question: not "is this data good or bad," but "what do I need to check before I use it." Introduce ALT/FNDATA briefly: a transaction-level pricing dataset for the secondary market in watches, jewelry, art, wine, automobiles, and other physical luxury assets, drawn from more than 100 auction houses and marketplaces.

10 to 20 min: Sandbox setup. Participants register at sandbox.altfndata.com with a work or school email address; approval is automatic. While waiting, briefly show the SQL editor, coverage browser, data dictionary, and export tools using a screen share. (For the 60-minute version, ask participants to register before the session and use this block only for a quick tour.)

20 to 35 min: Reading the coverage browser before touching a query. Hands-on, no code. Participants open the coverage browser tab and search for a company or brand of their choosing. They record: how many records exist for that brand, how many vendors report on it, and what time range is represented. Discuss as a group: does more coverage always mean better data? What might explain a brand with very few records?

35 to 55 min: Guided SQL exercise. A short primer on SELECT, WHERE, and ORDER BY (5 minutes), then participants paste and run the example query below in the sandbox SQL editor, modify the designer filter to a brand of their choice, and note what changes. (For the 60-minute version, skip the SQL primer and provide the query as a fill-in-the-blank template instead.)

SELECT designer, item_title, sale_date, usd_price_decimal
FROM all_watches_data
WHERE designer LIKE '%Omega%'
  AND status = 'sold'
ORDER BY sale_date DESC
LIMIT 25;

55 to 75 min: Coverage and provenance segment (the core data-literacy segment). This is the heart of the session. Walk participants through a real, responsibly framed example of a coverage gap: a continuously ingested transactional dataset like this one is still backfilling its most recent quarters as new auction results are processed, so a naive year-over-year comparison of the latest period against the prior one can look like a market decline when it is really a data-ingestion lag. Have participants run a query counting records by sale_date in recent months versus the same months a year earlier, and ask them to hypothesize what they are seeing before revealing the ingestion-lag explanation.

Discussion prompts for this segment:

  1. If a chart showed transaction volume dropping sharply in the most recent quarter across every category at once, what would that pattern itself tell you, regardless of the underlying cause?
  2. What is the difference between a dataset being wrong and a dataset being incomplete in a way that looks wrong if you do not check?
  3. What is one question you could have asked before trusting a year-over-year comparison from any dataset, not just this one?
  4. Stable metrics in this session, like total sold-lot counts, median realized price, or the ratio of realized price to pre-sale estimate, held up fine even where the year-over-year comparison did not. Why might an aggregate like a median be more resistant to a recency gap than a single period's total count?
  5. Where else, outside of auction data, have you seen a "most recent period looks worse" pattern that turned out to be a reporting or collection lag rather than a real change?

75 to 85 min: Small-group synthesis. In pairs or trios, participants pick a second brand or category and write down three things they would check before using this dataset in a paper: one about coverage, one about a specific field's definition, and one about the time period.

85 to 90 min: Wrap-up and take-home assignment. Recap the two general skills practiced: checking coverage before querying, and treating the most recent period of any live dataset with extra scrutiny. Assign the take-home exercise below.

Take-home exercise

Pick any two categories in the dataset (for example, handbags and fine art). For each, use the sandbox to find: the total count of sold lots, the median realized price, and the average pricing power (realized price divided by high estimate). Write three to five sentences comparing the two categories, and one sentence noting any coverage caveat a reader should know about before trusting your comparison (for example, an especially recent time window, or a category with very few vendors represented). Submit the exported CSV or a screenshot of the query results alongside the writeup.

Facilitator notes

  • This session works best after participants have registered in advance; live registration during the session is the most common source of delay.
  • No API key is required for this session. If a faculty member asks about deeper programmatic access for their own course, direct them to Access_and_Licensing_FAQ.md in this pack, or to info@altfndata.com.
  • The coverage and provenance segment is intentionally framed around a real, observable pattern in the dataset rather than a hypothetical one. Keep the tone matter-of-fact: this is normal behavior for any continuously ingested transactional dataset, not a flaw unique to this one, and naming it plainly is itself good modeling of the skill being taught.