I am thinking which bootstrap to choose for studying MIT-BIH arrythmia database (44 patients) in time-domain and Fourier-domain.
I need to find a variable which can allow me to determine
variance of the 1D data, but I am not sure about it.
I think bootstrap fits the data because there is only a few cases and independence of samples. The theoretical distribution is complicated and/or unknown. I need power calculations.
There are many events per time unit in the ECG signal, so I am thinking about Cluster Bootstrap.
I think this method has been used by some authors.
This R-bloggers post seems to be promising about the topic and here about speeding up the process.
Currently, I am using the default bootstrap, done as
data; data=datasample(data, size(data,1)); in Matlab.
My intuition about which bootstrap method good from best to worst
- Cluster bootstrap
- Time-series bootsrap
- Bayesian bootstrap
- Basic bootstap, like the one-liner in Matlab
- No bootstrap; having the cases in order
I take lastly a random sample from the group by
Testing the Cluster bootstrap from the first blog post
The only doubt is here that I need to find
stdev of the data, because I am doing cluster sampling so not sure which parameter, I should use to get the variables; there is entropy and other variables but not sure if anything standard.
Here doing bootstrapping from Normal distribution with mean=0, stdev=40 and sample size N=44
Loading required package: zoo Attaching package: ‘zoo’ The following objects are masked from ‘package:base’: as.Date, as.Date.numeric Loading required package: Matrix Loading required package: methods Attaching package: ‘lfe’ The following object is masked from ‘package:lmtest’: waldtest Call: lm(formula = y ~ x) Residuals: Min 1Q Median 3Q Max -73.767 -35.860 7.194 32.447 96.114 Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) 7.756 6.578 1.179 0.245 x 2.169 5.765 0.376 0.709 Residual standard error: 42.38 on 42 degrees of freedom Multiple R-squared: 0.00336, Adjusted R-squared: -0.02037 F-statistic: 0.1416 on 1 and 42 DF, p-value: 0.7086 Call: lm(formula = y ~ x, data = dat) Residuals: Min 1Q Median 3Q Max -73.767 -35.860 7.194 32.447 96.114 Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) 7.756 1.439 5.391 9.03e-08 *** x 2.169 1.261 1.720 0.0857 . --- Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Residual standard error: 41.45 on 878 degrees of freedom Multiple R-squared: 0.00336, Adjusted R-squared: 0.002225 F-statistic: 2.96 on 1 and 878 DF, p-value: 0.08571 Call: felm(formula = y ~ x | 0 | 0 | g, data = dat) Residuals: Min 1Q Median 3Q Max -73.767 -35.860 7.194 32.447 96.114 Coefficients: Estimate Cluster s.e. t value Pr(>|t|) (Intercept) 7.756 6.750 1.149 0.251 x 2.169 6.732 0.322 0.747 Residual standard error: 41.45 on 878 degrees of freedom Multiple R-squared(full model): 0.00336 Adjusted R-squared: 0.002225 Multiple R-squared(proj model): 0.00336 Adjusted R-squared: 0.002225 F-statistic(full model, *iid*): 2.96 on 1 and 878 DF, p-value: 0.08571 F-statistic(proj model): 0.1038 on 1 and 43 DF, p-value: 0.7488 Time difference of 0.1756198 secs  6.501451 Time difference of 3.301254 secs  6.501451
Which parameter can you use to get the variables
stdev in clustering the sample?
I see some colleagues in MathWorks doing 6th degree polynomial detrending in cardiac signals because of baseline shift i.e. due to respiratory signal.
However, there is little knowledge about its validity in physiological sense.
So some sort of detrending may help to get the
stdev of the phenomenon in the signal.
MIT-BIH signal has the frequency 180 Hz for 30 minutes recording. Doing the 6th degree detrending will put depolarisation of the signals to the specific interval so it can give us the desired
However, I have not found any publication which shows that the validity of the depolarisation can be maintained after the detrending.