There is no need to use a loop.
First, get used to using meaningful variable names. You have samplesize, and nsamples. What do they mean? I'll guess this:
nsim = 10000; % how many simulated runs? Here I used 10000 total simulated runs.
tossespersim = 100; % each simulated run has 100 coin tosses
There is no need to use binornd at all. But you can if you want to do so. Rand is simpler.
p = 0.5; % the coin is fair. p is the probability of a head
coins = (rand(nsim,tossespersim) <= p)*2 - 1; % -1 means we got a tail, 1 is a head.
Now just use mean to compute the means.
meantoss = mean(coins,2);
meantoss will be a vector of of length nsim.
size(meantoss)
The histogram is simple.
nbins = 10;
histfit(meantoss,nbins)
