The comments are right: you can't prove the complexity, and you need more points to model it empirically. This is shown by the plot of your current data.
If you get more data, one thing you could do to investigate it is a log-log plot, plotting the log of time t against the log of number of elements n. This gives you a straight line with gradient p if your relationship is of the form: t = n^p, as illustrated below with some simulated data. The black points are for a y=x^2 relationship, the green for y=x, and the red for y=x*log(x) is somewhere inbetween.
If you think it's an nlogn relationship, you could simply plot t against nlogn, and you should get a straight line. Regression analysis is certainly possible, but if you look at your current data, you actually get a very good straight line fit from a linear regression (shown on the first plot above), and I certainly wouldn't say that it's linear on that basis.