Two types of methods

To parameterise COP fluctuations, measures are needed which best characterise postural sway and detect differences. Actually, many postural sway measures exists but there is little common ground for selecting and interpreting these sures. However, agreement exists about the necessity to consider multiple mea-sures in order to get insight into the multifactorial nature of postural control. In this context, Harbourne et al. (2009) remarks that different measures taken to-gether offer a more comprehensive description of postural control with the ability

to understand specific characteristics in the system. In the last years, various re-search groups have shown that two groups of sway measures - referred to as linear and nonlinear methods due to their underlying models (Table 1.1 in Section 1.2) - have to be combined to allow for a more holistic view of the variability present in the postural control system (Duarte and Freitas, 2010; Harbourne et al., 2009;

Kirchner et al., 2012):

(a) Measures of the amount of variability, named global parameters (Table 1.1, left column).

(b) Measures of the temporal organisation of variability, named structural pa-rameters (Table 1.1, right column).

Parameters from group (a) are traditionally considered. They interpret all regular structure present in the signal. However, the underlying hypothesis of “variabil-ity is equivalent to white noise” is questionable. It includes the assumption that COP fluctuations are detrimental. But, postural sway can be exploratory as it generates information from the environment (Chagdes et al., 2009). At this point it has to be mentioned that not all noise is bad. There are different sorts of noise labelled with different colours e.g., white, pink, brown or black noise. The colour denotes the strength of the long-range correlations with black noise corresponding to a highly structured time series. One has to be specific when talking about noise in a system. Newell et al. (2006) indicate the failure of previous studies on noise and motor control, not to state explicitly what kind of noise it is referred to. Typ-ically, the noise interpretation of variability is related to white noise which means random occurrences. “The standard strategy has been to equate variability with noise without examining the type of noise and the structure of the variability”

(Newell et al., 2006, p.10). In contrast to a random occurrence, it could be shown that COP fluctuations contain meaningful structure (for review, see Stergiou and Decker, 2011). Newell (1998) already remarks that biological movement signals can be characterised by a variance profile other than white noise which indicates that error and variability are not synonymous. It is necessary to complement

COP parameterisation by a structural analysis in order to unravel hidden pat-terns in apparently random signals. Nonlinear methods aim at the identification of sub-units in the posturographic data which can then be related to the under-lying motor control processes. An overview of the structural parameters which are considered in the present work is given by Table 1.2. This choice of methods is based on literature references and on own research findings. More detailed information on the applied methods is given in Chapter 2. Basic formulas are presented there. In addition, we refer to the choice of input parameters as the results strongly depend on them. The present work should contribute to finding

Table 1.2. Overview of methods for the structural analysis of postural sway data which were considered in this thesis. The third column summarises exemplary literature concerning the application of the methods.

Method Meaning of interest Literature and Keshner (2008); Lake et al.

(2002); Ramdani et al. (2009, 2011); Rhea et al. (2011);

Roerdink et al. (2006); Stins et al. (2009)

Multi-scale entropy (MSE) based on SaEn (Costa et al., 2002, 2005)

Costa et al. (2003, 2007); Duarte and Sternad (2008); Kang et al.

(2009); Manor et al. (2010) et al. (2009); Deligni`eres et al.

(2011); Donker et al. (2007);

Doyle et al. (2005); Duarte and Sternad (2008); Duarte and

an answer to the problem of method selection and application. One example for a statistical method which aim at the reduction of the large amount of parameters extracted from COP time series is the principal component analysis (PCA). Up to date, only few studies followed this approach. So, recommendations are rare which is also due to the fact that the results seem to depend on the sample (Roc-chi et al., 2006: healthy vs. Parkinson’s disease subjects) and the task (Schubert et al., 2012a,b: single- vs. dual-task). In addition, the feature selection process mostly includes traditional parameters (Chiari et al., 2002; Rocchi et al., 2006).

This is not astonishing as the application of nonlinear methods is not straightfor-ward. Unknown pitfalls limit the interpretation and further processing. However, Harbourne et al. (2009) suggest that linear and nonlinear parameters provide dif-ferent information regarding postural control in sitting infants. Furthermore, it was shown that traditional measures do not correlate with the complexity index and that they load on different principal components (Kang et al., 2009; Manor et al., 2010). Several questions remain, inter alia, what is the practical usefulness of the methods, which parameters can discriminate between individuals, which one responds to a change of the task or is sensitive to improved stability, how has the raw data be processed and how have the methods be adequately applied.

The latter two aspects have to be considered at the beginning of the data analysis process. They strongly influence the results and in a next step the interpretations as well as our understanding of postural control. A central aspect is the sampling duration and frequency which is addressed in the following.