Filtering Ground Reaction Force Data from a Vertical Jump: The Art Form of Data Analysis

The way time-varying data are treated prior to analysis is a critical consideration when actionable decisions are made from results obtained from that data. So why do so many researchers ignore the need to treat ground reaction force (GRF) data collected during jump testing? Is it blissful ignorance or do they know something others do not?

The argument supporting data treatment (i.e., smoothing or filtering) for force and/or motion data is not new, and there is an abundance of literature on the topic. In short, data treatment is considered a crucial first step during data analysis because time-varying signals contain both desired information (the stuff we want to analyze) and unwanted information (noise from electrical interference, moving wires, cross-talk between device channels, etc.). These two types of information typically don't collaborate well because of the potential return of unstable and erroneous results. It's quite clear from the literature that motion (i.e., kinematics) data should be treated prior to analysis. It's also relatively clear that GRF data should be treated when combined with motion data for joint kinetic analyses. But, what about when GRF data are examined solely, particularly during a jump test? Do unstable and erroneous results occur?

The information to follow is a summary of a recent publication I wrote with some colleagues and current doctoral students (pictured above) in The Journal of Strength & Conditioning Research.

In that study, we collected bilateral GRF data during countermovement vertical jump (CMJ) tests performed by 21 skilled jumpers (NCAA Division 1 Men's Basketball Players). In short, we treated the recorded data using a common filtering method (low pass bi-directional Butterworth digital filter with a 50 Hz cutoff frequency) in addition to the same digital filter but with cutoff frequencies determined via more objective methods, specifically visual inspection of the frequency band (Graph A in the Figure Below) and the frequency under which 99% of the cumulative signal power was contained (Graph B in the Figure Below). These methods were compared to a no treatment condition (raw data) with respect to numerous CMJ performance and strategy variables commonly studied in the literature.

So Does Filtering Lead to Different Results?

Our results indicated that, yes, filtering produces different results compared to analyzing the raw data, but we argue that using objective methods to treat the data are not appropriate because the filter does not properly handle the nearly vertical slope of the GRF curve around the time of takeoff (See figure below). What this poor handling of the GRF curve near takeoff leads to is a delayed time of takeoff and therefore compromised performance variables, notably flight height (i.e., jump height), jump power, and explosiveness (modified reactive strength index). The reason for these compromised performance results relates to the way in which flight height is calculated because it is determined by the vertical velocity developed at the time of takeoff. As the velocity prior to takeoff decreases slightly as a consequence of the functional inability to continue applying force in triple extension, filtered data lead to extraction of a smaller vertical velocity value at the time of takeoff.

An interesting result that should not be overlooked was that none of the GRF or yank (i.e., rate of force development) variables were affected by the choice to filter data using any of the methods explored. We argue this result provides support for analyzing raw data during a CMJ test. I say this with hesitation as a long-time advocate for filtering GRF data during a jump test (apologies to all the authors who had me as a peer-reviewer!). However, there is good news for others who advocate for filtering the GRF data. For instance, we also found that no variable studied was different between the raw data method and the common practice filtering method of using a low pass bi-directional Butterworth digital filter with a 50 Hz cutoff frequency. This could be useful not only to appease the filter-freaks like myself, but also for CMJ trials that have some vertical GRF fluctuations during the standing period prior to starting the CMJ. The 50 Hz filter should adequately smooth those fluctuations without altering the start time of the CMJ nor any other variable of interest. If the standing period of the trial is too choppy for the 50 Hz filter, you might want to re-think your data collection process!