If the data is indeed flawed, then the numbers have to be taken with a huge grain of salt. It's a lot like survey data and much group design research done with small populations. With small n, something that's, say 90% accurate, could produce a wildly-inaccurate value. As the number of data increases, though, the data become more meaningful. All measurement systems that have wiggle room present gigantic problems with small n. It's how we occasionally get data which suggest that quackery cures back pain. Since our tools for assessing back pain are based on self-report and are thus not highly precise, and we have a sample of perhaps 12 people assigned to 2 groups, the lack of precision in the measurement may suggest that reiki alleviates pain by a statistically-significant margin. Meta-analysis, and large groups with solid controls, however, will give us a more accurate depiction of reality. It's not that measuring back pain by self report should be completely dismissed, but that you can't rule out fluctuations which are a result of the measurement system and have nothing to do with your independent variable with a small number of data.