Tuesday, 30 January 2018

The Evidence Based Straitjacket: You Don’t Need To Be Houdini To Escape!


Neil Gibson: Director of Sport, Performance and Health at Oriam: Scotland’s Sports Performance Centre

John Hill: Head of Sport Science at MK Dons FC

John Fitzpatrick: Sports Scientist at Newcastle United FC

Robert McCunn: Performance Sports Manager at Oriam: Scotland’s Sports Performance Centre


A recent article in Professional Strength & Conditioning entitled ‘The evidence based straitjacket’ by Prof. Ian Jeffries raised some interesting and thought provoking topics around the interaction between applied practice and academic research. We felt there was room for, in some cases, an alternative perspective to be presented.  What follows is our take on some of the key issues including experience, innovation, universities and the academic process.

We will start however by offering an alternative to the term ‘evidence-based’; if this term is seen as overly restrictive to the practitioner they may wish to consider using ‘evidence-informed’ as a proxy.  This subtle change in terminology perhaps offers a greater degree of freedom in their work philosophy by allowing the extant literature to guide their work without controlling it. 


This is a difficult entity to quantify; we have all heard the oft-stated question, do you have 10 years’ experience or 1 years’ experience repeated 10 times?  In contrast, what research allows practitioners to do, in a range of formats, is present their work, after reflection and self-evaluation, to a panel of their peers for critique and review.  This need not always be in an academic format, there are a number of blog’s, industry quarterlies and national governing body publications that allow us to share our ideas and, hopefully, provide the reader with new ones.  If, however, we rely on someone’s experience, be that time in the trade or number of organisations they have worked with it is difficult to assess how effective their practises were.  In lectures to students one should always be keen to differentiate between what their opinion (based on anecdote and experience) versus what they have objective data to support.  Perhaps this would be a useful tact to take, especially for those sharing information on social media.  A recent article on the newly formed ‘Sport Performance & Science Reports’ platform stresses the importance and usefulness of employing magnitude based inferences when we assess the change someone has made over time.  We would suggest that without such approaches, clearly steeped in the realm of research, it is difficult to assess how effective a coach has been, irrespective of their standing or perceived experience.  Thanks largely to the work of people like Will Hopkins and Martin Buchheit resources that help those (and we include ourselves in this) who are not necessarily statistically minded to employ their use are now publically available in easy to manage formats.  Even if we are not pursuing publication there is no reason why practitioners cannot adopt these research skills when interpreting data they have collected.

Information gathering of this nature does not always need to be quantitative in nature.  We would stress that one cannot ignore the perceptions of athletes and coaches who have experienced the benefit of effective and enjoyable training programmes; look good, feel good, play good is as true today as it was before the proliferation of research around strength and conditioning. As such, qualitative analysis through interviews and focus groups provides an alternate approach to support and validate the efficacy of their work.  As 007 and his quartermaster state, ‘youth is no guarantee of innovation and age is no guarantee of efficiency.’

A lack of innovation

Whilst it is difficult to generalise, we would argue that there still exists a high degree of innovation within the field of strength and conditioning. We have rarely, if ever, visited other clubs or institutions and left without learning something new or seeing a different way of implementing training concepts we were already familiar with.  We certainly don’t think that research, or, group think, has eroded this at all.  What may be stifling innovation and creativity, however, is the structure of departments that exist within sporting organisations. 

Many practitioners working in the industry started their career at a time when sport science departments were in their infancy and as such benefited from regular interactions with coaches, medical staff and other disciplines within sport.  This allowed many to make their own decisions, and mistakes, generating new knowledge from the information presented to them from a range of different sources.  It is worth pausing here; knowledge is information that is personalised, that we can interpret and link to the environment within which we find ourselves in.  Often this exists as tacit knowledge that we need to, somehow, present to the outside world by externalising it, in order for others to see our perspective and also, where appropriate, learn from. Prof. Jeffries also expressed this valid sentiment in the aforementioned article.

Unfortunately, as the industry has grown, so too have the departments which house the practitioners; in tandem, the opportunities for young practitioners to interact with other disciplines, athletes and coaches has diminished.  Individuals are recruited into roles that are niche and fairly narrow, we work in silos designated by our profession and as such are not able to assimilate information from a variety of sources, many of which are important to athletic development.  It is this, I suspect, that might be responsible for a perceived loss of innovation within the field, if indeed it exists at all.

Breeding grounds

We have heard from a number of sources that Universities need to do more to prepare students for high performance sport and perhaps this is true.  There is certainly a disconnect between what is taught and what is discussed within professional teams and national governing bodies regarding the biggest challenges they face.  Of course, this is true of many disciplines, especially those where the lecturers are no longer interacting regularly with the field or topic on which they are teaching.  In America, they have industry sabbaticals for academic staff who want to recalibrate their skill set to more closely mirror the requirements industry places on the students they teach.  This seems like a good idea and one that sport could adopt; we are sure there would be no shortage of academic staff who would value the opportunity to spend some time engaging with and working alongside practitioners at the sharp end of sport.

Despite this it is still academia that is proving to be the most fertile breeding ground for new and aspiring coaches, not to mention those further on in their career trajectory.  It would be difficult to find anyone in a position of power or influence within the realm of sport science or strength and conditioning that hasn’t, at some point, engaged with academia through the pursuit of a degree.  Whilst not every athlete who embarks on a career in sport will make it to the top level, similarly, not every sport student will end up in working in sport.  What an academic education does provide however is the ability to be critical in ones thought processes; to instil the ability to decide what might be appropriate for the athletes you are working with and what isn’t, based on the available evidence; to query and probe that which is presented as fact.  Far from being barriers to innovation, these are skills that ensure students are not overly influenced by conjecture and opinion but pursue lines of enquiry that provide useful and useable information about the interventions they choose.  The important question is never ‘show me what you do?’ it is ‘show me what you don’t do and why?’; the latter should reveal whether a sound rationale has been developed and why certain ways of working were, either through trial and error or desk based research, deemed inappropriate.

The academic process

For those of us with a foot in this camp publications are the currency we deal in and, along with other less formal channels of communication, are an effective way to share information (not knowledge, it doesn’t become that until someone reads it and internalises what they consider to be the key message(s)).  That is not to say that there are not issues.  Publication bias exists, so even when you read a review it is likely that the findings will be skewed in favour of positive outcomes (few journals will publish a study that reports no effect, unfortunately).  There is a disconnect between those writing the materials and those reading them; it is still fairly common for national governing bodies, national institutes and professional teams not to provide access to journals for their staff, the ones working directly with athletes.  This is perhaps why people such as Yann Le Meur have had such success with the approach of sharing studies via infographics, making them accessible to a larger audience. 

Things are changing, however, there are now numerous opportunities for postgraduate students to pursue Masters degrees and PhDs whilst working in and being funded by sport, closing the circle between academia and performance. Furthermore, the increase in the number of journals of the topic of strength and conditioning has allowed articles with different approaches to be published and shared with the public.  Universities are also becoming better at providing industry placements for their students so that they can combine their understanding of the research base with some practical skills gained on the job.  For this to grow, however, sporting organisations need to decide what value they place on the discipline and whether it can be a viable career for those who choose to pursue it.  For this to be the case there needs to be a more formal process of continued professional development that employers can view to determine just how ‘experienced’ potential applicants are. In the absence of such a scheme, publications provide a useful way of evidencing our own willingness to learn, develop new skills and share our information with a wider audience.

Another feature of the academic process is giving practitioners the skills and competencies to conduct research and decipher data. Key principles such as assessing reliability and validity of the tests practitioners are conducting on a daily basis is something that is often forgotten in the applied world. Furthermore, critically assessing the success of ones interventions is essential to provide best practice. Without the skills and knowledge to perform such assessments, practitioners are simply shooting in the dark. The academic process gives practitioners the ability to provide their own evidence informed practice.

Out of the straitjacket and into the water park

Rather than being a straitjacket, we see being research informed as more of a popular ride at water parks, ‘the rapids’.  From a relatively calm and spacious lagoon (where you generate the question) you move into the rapids (the research) where you are tossed around somewhat, often in an uncomfortable manner, before being ejected back into the lagoon knowing a little bit more (albeit not much) than before you started.  This is research; you channel your energy into answering a specific question, once you have addressed this you start the process again, however, in a more informed state.  Unlike the straitjacket you are never really free of it as there are always questions to be answered. It is the pursuit of these answers that allows us to be innovative and advance our field by creating shared knowledge through the information we create.

Monday, 25 August 2014

Pre-Match Salivary IgA in Soccer Players From the 2014 World Cup Qualifying Campaign


Purpose: To monitor resting salivary secretory immunoglobulin A (SIgA) levels in international soccer players during the short-term training period that precedes international match play.

Methods: In a repeated measure design, saliva samples were obtained from thirteen outfield soccer players who participated in the training camps preceding 7 games (5 home and 2 away) of the 2014 FIFA World Cup qualifying campaign. Samples were obtained daily for four days preceding each game (and analyzed for SIgA using the IPRO Oral Fluid Collection System) at match day (MD) minus 1 (MD-1), minus 2 (MD-2), minus 3 (MD-3) and minus 4 (MD-4).

Results: SIgA displayed a progressive decline (P = 0.01) during the 4-day training period (MD-4: 365 ± 127 μg.mL-1; MD-3: 348 ± 154 μg.mL-1; MD-2:  290 ± 138 μg.mL-1; MD-1:  256 ± 90 μg.mL-1) such that MD-1 values were significantly lower (P = 0.01) than both MD-4 and MD-3. Ninety-five % confidence intervals for the differences between MD-1 and MD-4 (95% CI = -191 to - 26) and MD-1 and MD-3 (95% CI = -155 to - 28).

Conclusions: Data demonstrate that a short-term soccer training camp in preparation for international competition induces detectable perturbations to mucosal immunity. Future studies should monitor SIgA (as a practical and non-invasive measurement of immunity) alongside internal and external measures of training load in an attempt to strategically individualize training and nutritional strategies that may support optimal preparation for high-level competition.

Sunday, 15 June 2014

Predictive Analytics in Microsoft Excel – Part 1: Correlation & Linear Regression

Before we start, you can find a brief introduction to this predictive analytics article series by clicking HERE.

Part 1 of the series is correlation & linear regression. When two variables are related to one another they are said to be correlated. A very simple example in a sporting context is player anthropometrics. When plotted on a scatter diagram the height and weight of your squad should correlate: i.e. taller people tend to weigh more.

If you can quantify the strength and direction of a correlation, you can then start to use your knowledge of one variable to predict another. For example, if you know the height of your player you can predict their weight. The process you use to assess the strength of a relationship is called correlation analysis; however the process you use to predict one variable from another is called regression analysis. The strength of your correlation analysis will determine the accuracy of your regression analysis.

To demonstrate correlation analysis and linear regression I will use something a little more interesting than height and weight. The majority of top-level teams will be using technology to monitor their players in training (GPS) and match play (Prozone). If your aim is to collate all data into one database, data taken from different sources needs to be accurate and comparable. We know from research (Harley et al., 2010) that caution must be taken when using these systems interchangeably, as GPS reported higher values for total distance covered (TD) during match play when compared to Prozone. One possible solution to allow the use of both data sets in one database is to create a linear regression model to predict GPS TD from Prozone match data.

In order to do this we need to evaluate the strength of the relationship (correlation) between GPS and Prozone TD data. It is important to firstly visualize the relationship using a scatter diagram. Below is a screenshot of what this should look like in Excel for Mac 2011. Please note this is simulated data, generated for the purpose of this article and does not accurately represent the true relationship between GPS and Prozone data.

Figure 1.1 Visualisation of the correlation between GPS and Prozone total distance match data.

When visualizing the data like this it is clear to see a positive, nearly perfect correlation (which is to be expected). In order to objectify this relationship we can calculate Pearson’s Correlation Coefficient. This can be easily done using Excel as long as your data is accurately input into a list, as shown in Figure 1.1.

To do this use the Excel formula =CORREL()

Figure 1.2 shows the correlation coefficient for our dataset is 0.98, this can be categorized as nearly perfect (for more information on categorizing the strength of your correlation see Will Hopkins – A New View of Statistics). To calculate this we used the formula =CORREL(A2:A31,B2:B31).

Figure 1.2 Correlation between GPD and Prozone data using =CORREL() function

So how do we get from a correlation to a prediction? In simple terms, when conducting a regression you forecast one variable from another. The equation is as follows:

The regression coefficient, b is obtained as follows:
  1. Find rxy, the correlation between the predictor variable x and the forecast variable y. 
  2. Multiple rxy by the standard deviation of the forecast variable y (Sy).
  3. Divide the result by the standard deviation of the predictor variable x (Sx).

Besides the regression coefficient you need to know the intercept, denoted as a, which you can calculate as follows:
  1. Multiple the regression coefficient b, by the mean of x.
  2. Subtract the result of step 1 from the mean of y.

So the equation for the intercept a is as follows:

a = y - bx

Figure 1.3 shows how this all comes together in an Excel worksheet.

Figure 1.3 Calculation of Linear Regression

Having calculated the regression coefficient and the intercept we can now forecast GPS TD based on Prozone TD. For Example:

y = bx + a

GPS TD = Regression Coefficient * Prozone TD + Intercept

GPS TD = E10 * 12000 + E11

This returns 12524. Given the observed relationship between Prozone and GPS total distance data, we can predict that 12000 metres derived from Prozone data would equal 12524 metres in GPS data.

Conducting this analysis manually helps to understand the underlying equations that are used in a linear regression and helps to show how the correlation is related to the regression formula. I would suggest trying to calculate your regression this way at least once to gain an understanding of these calculations. However, this is a waste of time for everyday work. Luckily there is an Excel function that can calculate this for you =LINEST(). Figure 1.4 shows how you can use this function on this data set.

Figure 1.4 Calculation of Linear Regression using =LINEST() function.

The =LINEST() function can be used to bypass all the intermediate calculations shown in Figure 1.3. This function calculates the regression coefficient and intercept directly. They appear in cells D2:E2 of Figure 1.4.

It must be noted the =LINEST() function is an Array Formula this means instead of simply pressing Enter, you hold down Ctrl and the Shift keys as you press Enter.

So in the case of our dataset in Figure 1.4, the sequence of events is as follows:
  1. Select cells D2:E2. 
  2. Type the formula =LINEST(A2:A31,B2:B31)
  3. Hold down Ctrl and Shift, and press Enter.

You should now see the regression coefficient appear in cell D2 and the intercept in E2. If you look in the formula box, you should see the array-entered formula surrounded by curled brackets, as shown in cell F2. Excel will automatically create these brackets when an array formula is entered you do not need to type these.

Just as in Figure 1.3, you can use the regression coefficient and intercept to create a forecast value based on a predictor value. The formula to do that is shown in cell F5.

Using this formula in cell E5, you can enter any Prozone TD data in cell D5 and a Predicted GPS TD will be returned in cell E5.

Excel does make all of this even easier, the =TREND() function puts everything into one step. It calculates the regression coefficient and intercept and applies them to one or more predictor values and displays a forecast value.

=TREND() will be covered in more detail later in this article series. I have not covered it in this section based on the guidance of Conrad Carlsberg in his book (See Introduction for a link). Conrad states, “By looking at the equations and some of the other results that pertain to the equations, you can do a better job of evaluating whether and how you want to use it.

That wraps up Part 1 of the Predictive Analytics article series. Part 2 will cover how to apply the techniques we have learnt to conduct multiple regression analysis, where you can use more than one predictor variable to return a forecast variable.

It would be good to hear how people would use this correlation and linear regression analysis in practice, please get in touch via the comments section or Twitter.

Click to download the Correlation & Linear Regression Workbook.