Data considerations

“If you have not found the fractal pattern, you have not taken enough data” – Machlup, 1977)

  • All analyses:
    • Data points: \(2^n\), minimum 1024 (\(2^{10}\))
    • Remove 3SD only if this is absolutely necessary
  • Spectral analysis:
    • Normalize the data (z-score transform: (X-mean(X))/SD(X)
    • Remove linear trend if necessary (detrend)
    • Decide number of frequencies to estimate, min. 512
  • SDA:
    • Normalize the data (z-score transform: (X-mean(X))/SD(X)
  • DFA:
    • Nothing extra, analysis integrates and detrends the signal