How bad data practice is leading to bad data research

https://www.forbes.com/sites/kalevleetaru/2018/02/19/how-bad-data-practice-is-leading-to-bad-research/#e76a25f1c354

A long but very well thought out article that encourages us to look beyond the ability to reuse data to whether the data is appropriate for  the questions being asked.

The very last paragraph sums it up:

Putting this all together, poor data practice, from honest statistical error to misunderstanding of data and methods, failure to normalize to malicious manipulation, coupled with copy-paste Google Scholar-ship, threatens to call into question many of the findings of data-driven research and is creating a dangerous landscape where a single honest spreadsheet error can reshape government policy, where "data" is conflated with "truth" and where dubious results are accepted as “fact” through the gilded veneer of data.

Towards a data literate citizenry - Article

Abstract:

We believe that data literacy should be a skill not just for scientists, but for all citizens. We make the case by considering literatures on various kinds of literacy, and using a number of examples to explore some of the challenges that emerge from trying to move to a more data literate citizenry. We consider the opportunities arising from developing technologies to help individual communities and intermediaries in understanding how data should inform personal and societal decision-making.

https://www.ideals.illinois.edu/handle/2142/38385

Inflammation Data Clash

Identical datasets yield opposite conclusions on the use of mice as models of human inflammation.

http://www.the-scientist.com//?articles.view/articleNo/40713/title/Inflammation-Data-Clash/

WIKIMEDIA, DOUG BECKERSIn 2013, a large group of collaborators published a paper in PNAS concluding that genomic responses to inflammatory stress in mice don’t correlate well with those in humans. “The prevailing assumption—that molecular results from current mouse models developed to mimic human diseases translate directly to human conditions—is challenged by our study,” Junhee Seok, who’s now at the Northwestern University Feinberg School of Medicine, and his colleagues wrote in their paper.

Fast forward to last week, and another study, using the very same data, reached the opposite conclusion.

“Here we re-evaluated the same gene expression datasets used in the previous study by focusing on genes whose expression levels were significantly changed in both humans and mice,” Keizo Takao and Tsoyosji Miyakawa wrote in their recent paper, also published in PNAS. “Contrary to the previous findings, the gene expression patterns in the mouse models showed extraordinarily significant correlations with those of the human conditions.”

So what gives?

According to a press release from Fujita Health University, where Miyakawa is based, the original study compared all the genomic changes, regardless of whether the involved genes only responded to the stress in one of the species. Such an approach “obscures the correlation between homologous genes of humans and mice to nearly zero, as demonstrated by Seok et al.” in the 2013 paper, the release read.

Gregory Hickok, a neuroscientist at the University of California, Irvine, tweeted: “Reflects that the conclusions aren’t automatically given by data. It’s all in the interpretation.”