Analysis of Parallel Spike Trains

Volume 7 of the series Springer Series in Computational Neuroscience pp 359-382

Generation and Selection of Surrogate Methods for Correlation Analysis

  • Sebastien LouisAffiliated withLaboratory for Statistical Neuroscience, RIKEN Brain Science Institute
  • , Christian BorgeltAffiliated withIntelligent Data Analysis and Graphical Models Research Unit, European Centre for Soft Computing
  • , Sonja GrünAffiliated withLaboratory for Statistical Neuroscience, RIKEN Brain Science Institute Email author 

* Final gross prices may vary according to local VAT.

Get Access


Generating artificial data from experimental data as a means for implementing a null hypothesis is becoming widely used. The reason is twofold: increasing computer power now allows for this type of approach, and it has become clear that the complexity of experimental data does not in general allow one to formulate a null hypothesis analytically. This is particularly true for the correlation analysis of parallel spike trains. Neglecting statistical features of experimental data can easily lead to the occurrence of false positive results, which of course needs to be avoided. Therefore surrogate data are used to generate the predictor by modifying the original data in such a way that the feature of interest (temporal coordination of spikes) is destroyed but other features of the data are preserved. The latter aspect is the most demanding and requires the selection of a surrogate type that best fits the data at hand. This chapter will demonstrate the need for such a selection and will show selection criteria.