3 ms·

Publishing more data behind our reporting

146 points by gballan 8y ago

gearhart 8y ago
This is a great move. Journalists are finally regaining our trust in a western world that really needs somebody to hold other institutions accountable. https://www.edelman.com/trust-barometer https://www.edelman.com/trust-barometer
MrEfficiency 8y ago
Recently my website was criticized for using the data on the food package rather than USDA data. I'm unsure what the 'correct answer' is, given they were off by 25% which is significant. I could add javascript so the user could decide which reporting method they would like to see, but I'm considering that this is massive overkill with limited benefit for the customer. I doubt my conclusions are impacted as despite this huge difference, top 10 items usually stay around the top 10. Any thoughts on how to make the data the most trustworthy and accurate?
- throwaway080383 8y ago
  Doesn't have to be fancy JavaScript. Even an asterisk next to disputed items and a link to the "Alternative Nutrition Facts" would be helpful.
  - MrEfficiency 8y ago
    Ive been thinking about this since the morning. I believe a link to the alternative calculation is probably going to be my solution.
- forapurpose 8y ago
  Is there any information available about why the two sources differ so much? For example, is the USDA data not updated frequently?
Sir_Cmpwn 8y ago
Unrelated to the article: why does a news organization with an established web property stoop to writing articles on Medium?
- rm_-rf_slash 8y ago
  Because non-subscribers can’t get past the third paragraph of the paywall?
devmacrile 8y ago
Publishing raw data itself is definitely a good start but there also needs to be a push towards a standardized way of sharing data along with it's lineage (dependent sources, experimental design/generation process, metadata, graph relationship of other uses, etc.).
- westurner 8y ago
  > Publishing raw data itself is definitely a good start but there also needs to be a push towards a standardized way of sharing data along with it's lineage (dependent sources, experimental design/generation process, metadata, graph relationship of other uses, etc.). Linked Data based on URIs is reusable. ( https://5stardata.info https://5stardata.info ) The Schema.org Health and Life Sciences extension is ahead of the game here, IMHO. MedicalObservationalStudy and MedicalTrial are subclasses of https://schema.org/MedicalStudy https://schema.org/MedicalStudy . {DoubleBlindedTrial, InternationalTrial, MultiCenterTrial, OpenTrial, PlaceboControlledTrial, RandomizedTrial, SingleBlindedTrial, SingleCenterTrial, and TripleBlindedTrial} are subclasses of schema.org/MedicalTrial. A schema.org/MedicalScholarlyArticle (a subclass of https://schema.org/ScholarlyArticle https://schema.org/ScholarlyArticle ) can have a https://schema.org/Dataset https://schema.org/Dataset. https://schema.org/hasPart https://schema.org/hasPart is the inverse of https://schema.org/isPartOf https://schema.org/isPartOf . More structured predicates which indicate the degree to which evidence supports/confirms or disproves current and other hypotheses (according to a particular Person or Persons on a given date and time; given a level of scrutiny of the given information) are needed. In regards to epistemology, there was some work on Fact Checking ( e.g. https://schema.org/ClaimReview https://schema.org/ClaimReview ) in recent times. To quote myself here, from https://news.ycombinator.com/item?id=15528824 https://news.ycombinator.com/item?id=15528824 : > In terms of verifying (or validating) subjective opinions, correlational observations, and inferences of causal relations; #LinkedMetaAnalyses of documents (notebooks) containing structured links to their data as premises would be ideal. Unfortunately, PDF is not very helpful in accomplishing that objective (in addition to being a terrible format for review with screen reader and mobile devices): I think HTML with RDFa (and/or CSVW JSONLD) is our best hope of making at least partially automated verification of meta analyses a reality. "#LinkedReproducibility"; "#LinkedMetaAnalyses", "#StudyGraph"

146 points by gballan 8y ago

This is a great move. Journalists are finally regaining our trust in a western world that really needs somebody to hold other institutions accountable. https://www.edelman.com/trust-barometer https://www.edelman.com/trust-barometer

MrEfficiency 8y ago

Recently my website was criticized for using the data on the food package rather than USDA data. I'm unsure what the 'correct answer' is, given they were off by 25% which is significant. I could add javascript so the user could decide which reporting method they would like to see, but I'm considering that this is massive overkill with limited benefit for the customer. I doubt my conclusions are impacted as despite this huge difference, top 10 items usually stay around the top 10. Any thoughts on how to make the data the most trustworthy and accurate?

throwaway080383 8y ago
Doesn't have to be fancy JavaScript. Even an asterisk next to disputed items and a link to the "Alternative Nutrition Facts" would be helpful.
- MrEfficiency 8y ago
  Ive been thinking about this since the morning. I believe a link to the alternative calculation is probably going to be my solution.
forapurpose 8y ago
Is there any information available about why the two sources differ so much? For example, is the USDA data not updated frequently?

Sir_Cmpwn 8y ago

Unrelated to the article: why does a news organization with an established web property stoop to writing articles on Medium?

rm_-rf_slash 8y ago
Because non-subscribers can’t get past the third paragraph of the paywall?

devmacrile 8y ago

Publishing raw data itself is definitely a good start but there also needs to be a push towards a standardized way of sharing data along with it's lineage (dependent sources, experimental design/generation process, metadata, graph relationship of other uses, etc.).

westurner 8y ago
> Publishing raw data itself is definitely a good start but there also needs to be a push towards a standardized way of sharing data along with it's lineage (dependent sources, experimental design/generation process, metadata, graph relationship of other uses, etc.). Linked Data based on URIs is reusable. ( https://5stardata.info https://5stardata.info ) The Schema.org Health and Life Sciences extension is ahead of the game here, IMHO. MedicalObservationalStudy and MedicalTrial are subclasses of https://schema.org/MedicalStudy https://schema.org/MedicalStudy . {DoubleBlindedTrial, InternationalTrial, MultiCenterTrial, OpenTrial, PlaceboControlledTrial, RandomizedTrial, SingleBlindedTrial, SingleCenterTrial, and TripleBlindedTrial} are subclasses of schema.org/MedicalTrial. A schema.org/MedicalScholarlyArticle (a subclass of https://schema.org/ScholarlyArticle https://schema.org/ScholarlyArticle ) can have a https://schema.org/Dataset https://schema.org/Dataset. https://schema.org/hasPart https://schema.org/hasPart is the inverse of https://schema.org/isPartOf https://schema.org/isPartOf . More structured predicates which indicate the degree to which evidence supports/confirms or disproves current and other hypotheses (according to a particular Person or Persons on a given date and time; given a level of scrutiny of the given information) are needed. In regards to epistemology, there was some work on Fact Checking ( e.g. https://schema.org/ClaimReview https://schema.org/ClaimReview ) in recent times. To quote myself here, from https://news.ycombinator.com/item?id=15528824 https://news.ycombinator.com/item?id=15528824 : > In terms of verifying (or validating) subjective opinions, correlational observations, and inferences of causal relations; #LinkedMetaAnalyses of documents (notebooks) containing structured links to their data as premises would be ideal. Unfortunately, PDF is not very helpful in accomplishing that objective (in addition to being a terrible format for review with screen reader and mobile devices): I think HTML with RDFa (and/or CSVW JSONLD) is our best hope of making at least partially automated verification of meta analyses a reality. "#LinkedReproducibility"; "#LinkedMetaAnalyses", "#StudyGraph"