This was likely your source: a paper in the open access open-access journal
PLoS ONE by Daniele Fanelli of the University of Edinburgh, and here is a relevant excerpt from that paper: (I underlined the conclusion)
"...
On average, across the surveys, around 2% of scientists admitted they had "fabricated" (made up), "falsified" or "altered" data to "improve the outcome" at least once, and up to 34% admitted to other questionable research practices including "failing to present data that contradict one's own previous research" and "dropping observations or data points from analyses based on a gut feeling that they were inaccurate."
In surveys that asked about the behaviour of colleagues, 14% knew someone who had fabricated, falsified or altered data, and up to 72% knew someone who had committed other questionable research practices.
In both kinds of surveys, misconduct was reported most frequently by medical and pharmacological researchers. This suggests that either the latter are more open and honest in their answers, or that frauds and bias are more frequent in their fields. The latter interpretation would support growing fears that industrial sponsorship is severely distorting scientific evidence to promote commercial treatments and drugs.
As in all surveys asking sensitive questions, it is likely that some respondents did not reply honestly, especially when asked about their own behaviour. Therefore, a frequency of 2% is probably a conservative estimate, while it remains unclear how the figure of 14% should be interpreted.
More information: Fanelli D (2009) How Many Scientists Fabricate and Falsify Research? A Systematic Review and Meta-Analysis of
Survey Data.
PLoS ONE 4(5): e5738. doi:10.1371/journal.pone.0005738,
http://dx.plos.org/10.1371/journal.pone.0005738 "