The problem with these studies is that they don't test the particular conditions people are worried about: long term low level exposure due to radiation inside the body and organs. Nuclear plant workers are the best benchmark we have for that, and sure enough rates of cancer are pretty consistently shown to be higher among that group.
That's not right.
1. The risks of internal radiation are reasonably well understood from observations of patients given injections of ‘Thorotrast’, Radium dial painters exposed to 226 Ra and 228 Ra & patients given 224 Ra for medical conditions, occupational exposure of uranium miners to radon, and residential radon exposure etc.
Harrison, J. & Day, P., 2008. Radiation doses and risks from internal emitters. Journal Of Radiological Protection, J. Radiol. Prot., 28, 137–159.
2. Nuclear plant workers consistently have much better health (e.g. a lower cancer rate) then the general population (McGeoghegan et al. 2008). It's called the Healthy Worker Effect, it's one of the reasons why detecting elevated risks amongst nuclear workers is difficult. However, the few exceptions are subset of rare blood cancers e.g. Leukaemia, possibly lung cancer and heart disease. Even then, the detection of a statistically significant increases are limited to a few workers in the highest dose categories, >100 mSv for rare types of cancers & >300 mSv for Heart Disease. Note that cancer risk varies ~40% due to non-radiation effects (lifestyle and environment) and overwhelm the subtle effects (if any) of very low level radiation (below 100 mSv).
"Even a study of millions of workers exposed to very low doses (below 100 mSv), no matter how carefully conducted, would be inadequate to produce precise and uncertain estimates of risk in part because of the dominating influence of any subtle biases or unknown confounding factors" (UNSCEAR 2008, Land 1980).
The last and biggest study into Nuclear workers, "The 15 country Study" by Cardis et al. 2005 tracked the heath of up to 407,391 workers over 5.2 million person years, was revealed to be flawed by erroneous data in the Canadian cohort (involving 11,907 workers). Cardis et al. claimed, oddly, that all cancers except Leukaemia were elevated due to radiation (and there didn't even seem to the a threshold!). But when the erroneous Canadian data was identified and removed, there was no statistically significant increase in cancer. This fiasco highlights the sensitivity to confounding, bias and error in studies of low statistical certainty. Furthermore, Boice 2010 explained that there may never be another study of nuclear workers of this size, because the permitted exposures to workers are now far more stringent and few workers will ever be exposed to >100 mSv (or even >300 mSv) to detect the effects of radiation with statistical certainty (even considering the recent events in Japan).
So we have no direct evidence that a dose below 100 mSv causes cancer, that's why laboratory studies are so important, where we can investigate cellular repair and defence mechanisms that may cause radiation risk to depart from LNT i.e. threshold or hormesis.
Ashmore, J. P., Gentner, N. E. & Osborne, R. V., 2010. Incomplete data on the Canadian cohort may have affected the results of the study by the International Agency for Research on Cancer on the radiogenic cancer risk among nuclear industry workers in 15 countries. Journal of Radiological Protection 30, 121–129.
Boice, J. D. 2010. Uncertainties in studies of low statistical power. J Radiol Prot 30, 115–120.
McGeoghegan, D., Binks, K., Gillies, M., Jones, S. & Whaley, S., 2008. The Non-Cancer Mortality Experience of Male Workers at British Nuclear Fuels Plc, 1946–2005. Int. J. Epidemiol. 37, 506–518.