MATLAB determines the upper limit of your for loop once, when the for loop starts. Changing what that expression will return (by deleting rows from your vector, for example) does not cause MATLAB to re-evaluate that expression each time through the loop.
As an aside, you shouldn't use length with a matrix. Use size and specify the dimension input to retrieve the size of the array in that dimension (dim = 1 for rows, dim = 2 for columns, dim = 3 for pages, etc.)
So you start off, say, with a matrix with 100 rows. When i = 3 you delete a row, so you now have a matrix with 99 rows. But your for loop is still going to count to 100, meaning at that last iteration you're going to ask for the 100th row of a 99 (or fewer) row matrix.
How can you avoid this problem? You could iterate your for loop backwards, starting with the last row and working your way to the front. You could create a vector indicating whether to keep or delete the row and use that after the loop has finished to perform the deletion.
Alternately you could skip the loop entirely. I'm going to create an example data matrix of size [10 7] whose elements are integer values between 1 and 20. I'm going to specify a seed value to make sure that this example demonstrates what I want it to demonstrate.
rng(1, 'twister')
rawDataMatrix = randi(20, 10, 7)
Now create a vector indicating where the third column of rawDataMatrix is greater than 10.
isGreaterThan10 = rawDataMatrix(:, 3) > 10
Now use that vector to delete the rows whose third column is greater than 10.
rawDataMatrix(isGreaterThan10, :) = []
You can do this without the intermediary variable if you want, but I wanted to show each step for clarity.
rng(1, 'twister')
rawDataMatrix = randi(20, 10, 7)
rawDataMatrix(rawDataMatrix(:, 3) > 10, :) = []
