improve statistic of multidimensional data

Hey,

I have some data point with multiple properties and i need to scale up those datapoints to a higher quantity while keeping the relativ occurance of all the properties.

So for example i have some data A with 5000 elements where the value of A describes the property a of all 5000 individuell elements, then i have B with 5000 elements which describes property b, and so on with c,d,e,f,g,.... i think for the start 3 property erray should be enough. The values of A ,B and C can be big or small maybe they will describe e bimodal behaviour in A and a gaussian in B and maybe lognormal in C.

I now want to scale up my data to lets say 20000 datapoint while keeping the properties of A, B and C in bimodal, gausion, lognormal shape.

I tried to get 3 binary histograms with histcounts2, respectively between A-B, B-C and A-C, smoothed them a little bit to fill out the "holes" in my 3 matrices since they come from the too little statistics of only those 5000 datapoints, and then tried to reconstruct the 3d matrix form the 3 2D matrices, but somhow it didnt work.

I also tried:

https://de.mathworks.com/matlabcentral/answers/293645-how-to-sort-3d-data-into-bins

but this didnt work neither.

Does anybody have an idea how to do this properly??

Many thanks in advance and best regards

[John D'Errico]

Lets see. You want to ue statistics to improve the statistics of your data? Why do I feel this breaks some law? At least a law of reason. Essentially, what you seem to be describing is something I would call statistical impropriety.

You cannot use statistics to improve your data to then be able to make better statistical inferences from your data. If you could do this, then logically, you could them inprove your data again, and again, until eventually you would be able to make any conclusion you want, with near 100% certainty. Yes, this is done by politicians worldwide, on a daily basis. (And you can probably guess my opinion of such an activity.)

[Marc Laub]

So lets put it another way.

Its a about a simulation of particel growth where some particles growth at cost of other particles and they will growth untill some equilibrium is reached.

The problem is now that that equilibrium would be at lets say size x, but in the simulation the equilibrium oly reaches x/2 because the maximum size the particles can reach is limited by the start volume.

So i could just start the simulation with a larger start volume, i.e amore particles but this is at cost of simulation time which i do not want.

So what i wanna do is to increase my simulated volume from time to time and the additional value should have the same properties that my volume had at this point of time.

So the "fiiling the holes" which i described earlier is some szenario where i add particles to my simulation to increase the total volume, from which i know that the possibility of finding these particles whith their properties would be very likely if the observed volume was larger.

So its more like and interpolation of data from which we know that we would observe them with very high propability if the volume was large enough.

Hope this puts it in another context.

Its like if we have 5 particles with the size [1 1 2 3 4 5 ] and a shape factor of [0.9 0.8 0.8 0.8 0.7] and we would add a particle of size 1 with a shape factor of 0.85. And this is only done when I am sure that the quantity of particle I have in my simulation is large anough to describe my whole system (unlike like in this case).