Evidencebased dentistry has long been defined as the only true way to answer a specific clinical question or solve a problem. Yet surprisingly it isn't without its flaws. Under normal circumstances, establishing 'best evidence' involves the collection and analysis of scientific data which is often represented in statistical form. Conclusions are then drawn from those statistics which make up the basis of reasoning.
Sounds pretty straightforward right? Well yes, until you consider the following… It's thought that 4070% of all clinical papers written within the medical field (dentistry included) contain one or more statistical errors.
To back this up, in 2011 Kim et al carried out a series of tests on 418 clinical papers in the field of dental research written between 1995 and 2009. In 111 cases the use of statistics was deemed inappropriate due to insufficient information. What's more, of the remaining 307 cases, it was found that statistical data was misused in 51.5% of them.
To put it another way.... out of 418 clinical papers, 262 of them were deemed serious enough to have led to misleading conclusions!
"Statistical methodology in oral and dental research: pitfalls and recommendations. Hannigan A, Lynch CD."
The answer may lie in the initial testing itself.... In many cases, observational studies simply 'happen' rather than be designed. For this reason, any data used is often intended for another purpose. This can result in skewed findings.
Other issues occur when observational studies have gone on for many years. Consequently, subsequent assessors may use differing degrees of measurement. This again can result in flawed statistics. Finally, inappropriate use of controls from highrisk groups is often used to generalise results. For all of these reasons above, a justification of the sample size is highly important. While you might feel that sample justification is primarily an exercise in 'covering your back' check out these findings by Lucena et al...
In 2011, they researched 226 clinical papers on the study of microleakage in Operative Dentistry and of those, just 1% justified their sample size. Naturally, as design flaws cannot be fixed in the analysis stage of the research, in some cases this led to misconceptions.
That said, failure to justify sample sizes isn't the only pitfall when carrying out clinical testing. There are other reasons for statistical errors including:
What's more, they occur in some of the most common statistical techniques such as:
DESCRIPTIVE STATISTICS 

Issues  Counteractions 
Failure to take into account so called 'irrelevant' information.

Consider that extreme outliers may extort the true value of the mean. Therefore comparisons should always be made between the mean and median before analysing results.

HYPOTHESIS TESTING 

Issues  Counteractions 
The heavy reliance on probability or pvalue – Vähänikkilä et al reported that 81% of 928 articles reviewed in four high impact dental journals reported Pvalues.

Look to include confidence intervals. These are still heavily underused in dental research but can give a better indication that the true hypothesis lies between an indicated range. (See the study carried out by Lehmann et al). To back this up, Kim et al reported that only 20 of the 307 dental papers reviewed in journals contained confidence intervals. In other words, they add more 'weight' to your findings.

SURVIVAL ANALYSIS 

Issues  Counteractions 
Survival analysis testing is often carried out retrospectively. When this happens it's impossible to take into account censoring problems such as patients dropping out of the study  all of which can affect the end results.

Survival analysis should always be carried out using cohorts of people from a particular starting point moving forwards in time. This way any censoring can be noted and included in the final results.

CORRELATIONS AND REGRESSIONS 

Issues  Counteractions 
Misuse of the Pearson's Correlation Coefficient. While it's used to great effect to measure the strength of an association between two continuous variables, many clinical assessors have wrongly used it to correlate ordinal data (e.g. questions answered on a scale of 15).

For ordinal data testing the preferred method should be Spearman's Correlation Coefficient. In addition, don't be swayed by variable selection methods. Instead, biological and clinical know how should always win the day. 
While evidencebased testing continues to be the number one factor in solving important dental issues, it can and does have its pitfalls. The key is understanding that they exist and more importantly, knowing what to do to avoid them.
Kim JS, Kim DK, Hong SJ. Assessment of errors and misused statistics in dental research. International Dental Journal
2011;61:163–7.
Lucena C, Lopez JM, Abalos C, Robles V, Pulgar R. Statistical errors in microleakage studies in operative dentistry. A survey of the literature 2001–2009. European Journal of Oral Sciences 2011;119:504–10.
Vahanikkila ̈ H, Nieminen P, Miettunen J, Larmas M. Use of statistical methods in dental research: comparison of four dental journals during a 10year period. Acta Odontologica Scandinavica 2011;67:206–11.
Tufte ER. The visual display of quantitative information. Cheshire, CT: Graphics Press; 1983.
Lehmann KM, Igiel C, Schmidtmann I, Scheller H. Four colormeasuring devices compared with a
spectrophotometric reference system. Journal of Dentistry 2010;38S:e65–70.
FernandesTaylor S, Hyun JK, Reeder RN, Harris AHS. Common statistical and research design problems in manuscripts submitted to high impact medical journals. BMC Research Notes 2011;4:304.
During the early 1970's while Brånemark was deep into his research into osseointegration, other Europeans followed suit. They included André Schröeder at Switzerland's Berne University who was working on a similar implant for clinical application in conjunction with the renowned Straumann Institute.
Longterm followup study of osseointegrated implants in the treatment of totally edentulous jaws.Adell R, Eriksson B, Lekholm U, Brånemark Pl, Jemt T. Int J Oral Maxillofac Implants 1990 Winter;5(4):34759. PMID: 2094653
1982 was a huge turning point for Brånemark and it was a point that he had been working towards for the past 17 years. This included 15 years of clinical followup trials. Yet despite the hugely positive results, he was reluctant to present his findings to the public at the Toronto conference, because he felt quite simply, that the world still wasn't ready for dental implants.
A 15year study of osseointegrated implants in the treatment of the edentulous jaw. Adell R, Lekholm U, Rockler B, Brånemark PI. Int J Oral Surg. 1981 Dec;10(6):387416. PMID: 6809663
Anyone who knows anything about dental implants is aware of the story of PerIngvar Brånemark and how he accidentally discovered the process he later named as osseointegration. You're probably also aware that despite him being named as one of the most influential people in the dental profession, he wasn't actually a dentist.
Osseointegration and its experimental background., Brånemark PI., J Prosthet Dent. 1983 Sep;50(3):399410. PMID: 6352924