Prompt #2 at Biodiversity Next workshop: Top 3-5 ways to improve citizen science data quality

[eellwood]

[Context: At the biodiversity_next conference workshop, we asked participants to come up with a list of 3-5 ways we can improve citizen science data quality and/or perceptions of cs data quality. These are the responses, separated by group.]

(Summary (https://github.com/tdwg/citizen-science/projects))

---

Look at project design:
-Don't ask people to record everything. Start off with a limited number of easily identifiable species, for ex.
-Guides should be simple
-Allow for recording of uncertainty in identities or higher level identification
-Good guides should give an indication of what is needed to identify that species, e.g., take a photo of both top and bottom sides of the insect
-Include notes that ask "why you think it is that species?", e.g., did you go by it's sound, what guides did you use, did you consider related or look-alike species?

---

User-assigned confidence, e.g., not at all, a bit, very certain; with allowed defaults. Greater usability but greater error. Doesn't improve quality, but help quantify it.

---

• Reduce misidentifications
• stimulate use of scientific names
• rank the participants more aware of the importance of data accuracy
• educate people before letting them participate (in a fun way, e.g., using a quiz where they have to validate species images. In this way you also learn about their level and learn which mistakes they make most.

---

• Utilizing picklist/controlled vocabulary
• Second pass for specimens done by one-off, infrequent users
• vetting training material
• Multiple pass data cleaning workflow, e.g., first pass with open refine to catch typos, then pass along to expert

---

• Adding uncertainty. Measure (%, slider, etc.). Some users won't add observation if they are not 100% sure, others will add if they are 50% sure
• More feedback to user
• Cross-checks, random validation by many users
• New ways of gamification, rewards

---

• Bias in space. Communicate bias and suggest/nudge people's recording
• Bias in taxonomic identification. Guides, e.g., in app, keys, maybe AI, expert review
• Bias in habitat classification (as above)
• Bias in effort. Ask user, e.g., complete list
• Who is advertising: Data providers with a vested interest in reuse or (in the future) those whose funders moderate it to ensure openness.
• Who is using: Any modeler or data user should use this, but maybe we need to raise awareness and use of quality metadata

---

Collection manager --> advertise quality
Data user --> buy-in

---

• Species identification. Citizens shouldn't be forced to choose if they don't know
• Understanding data use
• Form layout/instructions, need to understand how to communicate

---

• Create applications w/ constraints and guides for data collection. Do quality assurance on the front end. But leave a free text field for remarks, even if not scientifically useful.
• Create age- and education-level appropriate training materials that have been tested and vetted by the user community as being easy to use and understand, and interesting.
• Introduce the concept of uncertainty - have users calibrate themselves/their data and report some level of uncertainty

---

• Software that insulates the citizen from the technical details
• Better matching of project managers with solid project design and the citizen scientists interested in and capable of doing it
• personalization /ownership of outcome

---

• Better match the protocol, the app, and the audience
• Better mechanisms to keep people motivated and provide good quality data
• Better mechanisms to validate/curate data

---

• Cost of validation based on automated task assignment to the most well adapted citizen scientists
• Improved data quality on site, based on automated feedback to the participant (e.g., "This species is not frequent here")
• Implement evaluation of participant profiles in order to detect wrong patterns of platform images