SPOTTING BAD SCIENCE

1. SENSATIONALISED HEADLINES Headlines of articles are commonly designed to entice viewers into clicking on and reading the article. At best, they over-simplify the findings of research. At worst, they sensationalise and mis­ represent them .

2. MISINTERPRETED RESULTS News articles sometimes distort or misinterpret the findings of research for the sake of a good story, intentionally or otherwise. If possible, try to read the original research, rather than relying on the article based on it for information.

3. CONFLICT OF INTERESTS Ma ny companies employ scientists to carry out and publish research - whilst this does not necessarily invalidate research, it should be analysed with this in mind. Resea rch can also be misrepresented for personal or financial gain.

4. CORRELATION Et CAUSATION Be wary of confusion of correlation & causation. Correlation between two variables doesn't automatically mean one causes the other. Global warming has increased since the 1800s, and pirate numbers decreased, but lack of pirates doesn't cause global warming.

5. SPECULATIVE LANGUAGE ()

Speculations from research are just that - speculation. Be on the look out for words such as 'may', 'cou ld', 'm ight', and others, as it is unlikely the research provides hard evidence for any conclusions they precede.

6. SAMPLE SIZE TOO SMALL In trials, the smaller a sample size, the lower the confidence in the results from that sample. Conclusions drawn should be considered with this in mind, though in some cases small samples are unavoidable. It may be cause for suspicion if a large sample was possible but avoided.

7. UNREPRESENTATIVE SAMPLES In human trials, researchers will try to select individuals that are representative of a larger population. If the sample is different from the population as a whole, then the conclusions may well also be different.

8. NO CONTROL GROUP USED In clinical trials, results from test subjects should be compared to a 'control group' not given the substance being tested. Groups should also be allocated randomly. In general experiments, a control test should be used where all variables are controlled.

9. NO BLIND TESTING USED To prevent any bias, subjects should not kn ow if they are in the test orthe control group. In double­ blin d testing, even researchers don 't know which group subjects are in un til after testing. Note, blin d testing isn't always feasible, or ethical.

10. 'CHERRY-PICKED' RESULTS This involves selecting data from experiments which supports the conclusion of the research, whilst ignoring those that do not. If a research paper draws conclusions from a selection of its resu lts, not all, it may be cherry-picking.

11. UNREPLICABLE RESULTS Results should be replicable by in depen dent research, an d tested over a wide range of cond itions (where possible) to ensure they are generalisable. Extraordinary claims require extraordinary evi dence - that is, much more than one in depen dent study!

12. JOURNALS Et CITATIONS Resea rch published to major journals will have undergone a review process, but can still be flawed, so should still be evaluated with these points in mind. Similarly, large numbers of citations do not always indicate that research is highly regarded .

• 2014 COMPOUND INTEREST - WWW.COMPOUNDCHEM.COM L@l<[92+W,J