[from Science News 187(2):24f., Jan. 24, 2015; published as a sidebar to Saey’s article “Repeat Performance: Too many studies, when replicated, fail to pass muster“]

12 reasons research goes wrong

Barriers to research replication are based largely in a scientific culture that pits researchers against each other in competition for scarce resources. Any or all of the factors below, plus others, may combine to skew results. 

Pressure to publish

Research funds are tighter than ever and good positions are hard to come by. To get grants and jobs, scientists need to publish, preferably in big-name journals. That pressure may lead researchers to publish many low-quality studies instead of aiming for a smaller number of well-done studies. To convince administrators and grant reviewers of the worthiness of their work, scientists have to be cheerleaders for their research; they may not be as critical of their results as they should be.

Impact factor mania

For scientists, publishing in a top journal — such as Nature, Science or Cell — with high citation rates or “impact factors” is like winning a medal. Universities and funding agencies award jobs and money disproportionately to researchers who publish in these journals. Many researchers say the science in those journals isn’t better than studies published elsewhere, it’s just splashier and tends not to reflect the messy reality of real-world data. Mania linked to publishing in high-impact journals may encourage researchers to do just about anything to publish there, sacrificing the quality of their science as a result.

Tainted cultures

Experiments can get contaminated and cells and animals may not be as advertised. In hundreds of instances since the 1960s, researchers misidentified cells they were working with. Contamination led to the erroneous report that the XMRV virus causes chronic fatigue syndrome, and a recent report suggests that bacterial DNA in lab reagents can interfere with microbiome studies.

Bad math

Do the wrong kinds of statistical analyses and results may be skewed. Some researchers accuse colleagues of “p-hacking,” massaging data to achieve particular statistical criteria. Small sample sizes and improper randomization or “blinding” of the researchers can also lead to statistical errors. Data-heavy studies require multiple convoluted steps to analyze, with lots of opportunity for error. Researchers can often find patterns in their mounds of data that have no biological meaning.

Sins of omission

To thwart their competition, some scientists may leave out important details. One study found that 54 percent of research papers fail to properly identify resources, such as the strain of animals or types of reagents or antibodies used in the experiments. Intentional or not, the result is the same: Other researchers can’t replicate the results.

Biology is messy

Variability among and between people, animal and cells means that researchers never get exactly the same answer twice. Unknown variables abound and make replicating in the life and social sciences extremely difficult.

Peer review doesn’t work

Peer reviewers are experts in their field who evaluate research manuscripts and determine whether the science is strong enough to be published in a journal. A sting conducted by Science found some journals that don’t bother with peer review, or use a rubber stamp review process. Another study found that peer reviewers aren’t very good at spotting errors in papers. A high-profile case of misconduct concerning stem cells revealed that even when reviewers do spot fatal flaws, journals sometimes ignore the recommendations and publish anyway (SN: 12/17/14, p. 25).

Some scientists don’t share

Collecting data is hard work and some scientists see a competitive advantage to not sharing their raw data. But selfishness also makes it impossible to replicate many analyses, especially those involving expensive clinical trials or massive amounts of data.

Research never reported

Journals want new findings, not repeats or second-place finishers. That gives researchers little incentive to check previously published work or to try to publish those findings if they do. False findings go unchallenged and negative results — ones that show no evidence to support the scientist’s hypothesis — are rarely published. Some people fear that scientists may leave out important, correct results that don’t fit a given hypothesis and publish only experiments that do.

Poor training produces sloppy scientists

Some researchers complain that young scientists aren’t getting proper training to conduct rigorous work and to critically evaluate their own and others’ studies.

Mistakes happen

Scientists are human, and therefore, fallible. Of 423 papers retracted due to honest error between 1979 and 2011, more than half were pulled because of mistakes, such as measuring a drug incorrectly.

Fraud

Researchers who make up data or manipulate it produce results no one can replicate. However, fraud is responsible for only a tiny fraction of results that can’t be replicated.