Added: Please note that the following discussions are based on my background in probability theory. In your example, binomial probabilities are defined only when \x*\n are integers; that is, these probabilities are defined only when \x is equal to one of the following 16 numbers
0, 1/15, 2/15, ... , 14/15, 15/15
Thus, in order to smoothly extend the graph, you have many options. These extensions include but are not limited to
Analytical continuation via the gamma function. But, you should be aware of the fact that
TeX is no Mathematica or MATLAB or Maple
and therefore the gamma function is not available on the fly. So, @marmot provided an excellent answer in which the gamma function is hardcoded using its asymptotic expansions. Because it uses asymptotic expansions, there could be unexpected computation errors (but only if you choose extreme domain to be plotted).
Use smooth in combination of the appropriate sample points. The smooth key basically interpolates between two points using splines. Usually, a twice continuously differentiable spline (a cubic spline) is used and it is good enough to fool the human eyes. This is what I propose in my new answer.
Use Gaussian or normal density approximation. There is a Local Central Limit Theorem which states
The discrete binomial probabilities can be approximated by a continuous Gaussian/normal density curve: The larger the \n*\p and \n*(1-\p), the better the approximation. (cf. Probability: Theory and Examples (4th ed.) by Rick Durrett, Sections 3.1 and 3.5).
This is what I propose in my old answer.
Please note that care must be taken when doing factorial calculations! If you pass real numbers into binom, then be prepared to get a “probability” bigger than one (cf. this question of mine). In your original graph, some “probabilities” seems to be bigger than two!
New answer
I just realized the OP wanted a “normalized binomial probability mass plot” instead of a “normal density approximation plot”. In this case, you can tell TikZ to draw only the points at which \x*\n are integers by specifying samples=\n+1. Remember to use integers for factorial calculations and to normalize in the coordinates afterward.
\documentclass{article}
\usepackage{pgfplots}
\pgfplotsset{compat=1.7}
% https://tex.stackexchange.com/questions/198572/tikz-binomial-distribution
\begin{document}
\begin{tikzpicture}[
declare function={binom(\x,\n,\p)=\n!/(\x!(\n-\x)!)\p^\x(1-\p)^(\n-\x);},
% declare function={normd(\x,\n,\p)=binom(\x\n,\n,\p);}% <- This is bad practice
]
\begin{axis}
\addplot[cyan, samples at={0,1,...,15}, smooth](x/15,{binom(x, 15, 0.7)});
\end{axis}
\end{tikzpicture}
\end{document}
Notice the top of the curve is around 0.2. Indeed, binom(11,15,0.7) = 0.218623131...
Old answer
Here, let’s use “Gaussian/normal approximation”. The approximating Gaussian/normal density takes the form
1/sqrt(2 * pi * n * p * (1-p)) * exp( - n * (x - p)^2 / (2 * p * (1-p)) )
Of course, this approximation works better if n*p and n*(1-p) are larger. If you are illustrating a normal approximation, then this would be your choice:
\documentclass{article}
\usepackage{pgfplots}
\pgfplotsset{compat=1.7}
% https://tex.stackexchange.com/questions/198572/tikz-binomial-distribution
\begin{document}
\begin{tikzpicture}[
declare function={binom(\x,\n,\p)=\n!/(\x!(\n-\x)!)\p^\x(1-\p)^(\n-\x);},
% declare function={normd(\x,\n,\p)=binom(\x\n,\n,\p);}% <- This is bad practice
declare function={normaldensity(\x,\n,\p)=exp(-\n(\x-\p)^2/(2\p(1-\p)))/sqrt(2pi\n\p*(1-\p));}
]
\begin{axis}
\addplot[cyan, samples at={0,1,...,15}, smooth](x/15,{binom(x, 15, 0.7)});
\addplot[orange, domain=0:1, smooth]{normaldensity(x, 15, 0.7)};
\end{axis}
\end{tikzpicture}
\end{document}
Here, the cyan curve is drawn using my new answer, and the orange curve is drawn using normal approximation.
smoothto the plot. However, then the plot is no longer accurate, i.e. the interpolation will cause the plot to deviate from the true values (a bit). I guess that the problem is that factorial is only defined for integers. You may want to use the Gamma function for a continuous interpolation. – Sep 05 '18 at 22:27