Mohie El-Din and Amein (2011) define a distribution in formula (1.2) which they call the exponential Bernoulli distribution (EBD). The distribution has the following form: $$\displaystyle f \left(t \right) = \left(1-p \right)~\alpha ~e^{-\alpha ~t }+p~\left(\alpha +\beta \right)~e^{-\left(\alpha +\beta \right)~t }$$ or with $\lambda = \alpha+\beta$ the distribution can also be written as : $$\displaystyle f \left(t \right) = \left(1-p \right)~\alpha ~e^{-\alpha ~t }+p~\lambda ~e^{-\lambda ~t }$$ In Mohie El-Din and Amein (2011) these densities are written in $x$. The logic behind this distribution is as follows. The Bernoulli variable $X$ with $X = 0, 1$ has a probability distribution with $P (X = 1) = p$ and $P (X = 0) = (1-p)$. Furthermore, there is a random variable $T$ with $0 \leq T$. If $X = 1$, then $T $ has an exponential distribution with rate parameter $\lambda$. If $X = 0$, then $T $ has an exponential distribution with rate parameter $\alpha$.

Now I have another case. In my case the following applies. If $X = 1$ then $T$ also has an exponential distribution: $$\displaystyle f \left(t \right) = p ~\lambda ~e^{-\lambda ~t }$$ But if $X = 0$, then $T = 0$. The final distribution is a proper probability distribution because: $$\displaystyle 1-p+\int_{0}^{\infty }\!p \, \lambda\,{{\rm e}^{-\lambda\,t}}\,{\rm d}t=1-p+p\int_{0}^{\infty }\!\lambda\,{{\rm e}^{-\lambda\,t}}\,{\rm d}t \, = \, 1$$ The expectation of $T$ can be written as: $$\displaystyle \mu_{{1}}\, = 0 \, (1-p) \, + \, \int_{0}^{\infty }\!p \, t \, \lambda\,{{\rm e}^{-\lambda\,t}}\,{\rm d}t \,= \, \frac {p}{\lambda}$$ The variance of $T$ can be written as: $$\displaystyle \mu_{{2}}\, = \, (1-p) \, \left(0-\frac{p}{\lambda}\right) ^2 +\int_{0}^{\infty }\!p \, \left( t-{\frac {p}{\lambda}} \right) ^{2}\lambda\,{{\rm e}^{-\lambda\,t}}\,{\rm d}t \, = \,-{\frac {p \left( p-2 \right) }{{\lambda}^{2}}}$$ The third central moment can be written as: $$\displaystyle \mu_{{3}}\, = \, (1-p) \, \left(0-\frac{p}{\lambda}\right) ^3 +\int_{0}^{\infty }\!p \, \left( t-{\frac {p}{\lambda}} \right) ^{3}\lambda\,{{\rm e}^{-\lambda\,t}}\,{\rm d}t \, = \,2\,{\frac {p \left( {p}^{2}-3\,p+3 \right) }{{\lambda}^{3}}}$$ I now have three questions. The first question is: what is the best way to name this distribution? The second question is: how can I best write down this distribution in a formula? The third question is: are the derivations of the three given central moments correct?


Mohie El-Din, M. M. and Amein, M. M. (2011). Estimation of Parameters of the Exponential Bernoulli Distribution Based on Progressively Censored Data. Applied Mathematical Sciences, Vol. 5, no. 58, 2883 - 2890.

  • $\begingroup$ 1. the authors gave a name to this distribution, what's wrong with "exponential bernoulli distribution"? 2. the formula for this distribution is given in your post via the density function $f(t)$. or do you mean something like $X\sim EBD(p, \alpha, \lambda)$ 3. Did you try solving the given integrals? $\endgroup$
    – Edgar
    Nov 14, 2019 at 16:10
  • $\begingroup$ Dear Edgar, My questions relate to the distribution that I propose, not to the distribution of Mohie El-Din and Amein. $\endgroup$ Nov 15, 2019 at 9:12
  • $\begingroup$ ah i see! what about "semi-exponential bernoulli distribution" :) $\endgroup$
    – Edgar
    Nov 15, 2019 at 9:18
  • $\begingroup$ Dear Edgar, That would be great. What about my other two questions? $\endgroup$ Nov 15, 2019 at 14:52

1 Answer 1


I looked into the paper and they define the EBD density function for $\alpha, \beta \geq 0$, which would include your case.

But for $\alpha \rightarrow 0$ the right tail of the distribution gets fatter. The mean of the EBD is $\frac{p}{\alpha+\beta} + \frac{1-p}{\alpha}$, which obviously doesn't exist for $\alpha=0$.

Notice that you can always give the cumulative distribution function for the EBD (see the paper you mentioned), but you cannot define a distribution density function for the case $\alpha=0$. Even though $\lim_{\alpha\rightarrow0}\alpha e^{-\alpha t}=0$, the resulting $\lim_{\alpha \rightarrow 0} f(t) = 0 + p (0+\beta)e^{-(0+\beta)t} = p\beta e^{-\beta t}$ is not a density function (integral is not equal to $1$).

You can still calculate the moments with the cdf (see here), though! Which gives the same central moments like your calculations.

Reagrding the EBD, see also a similar question here.

  • 1
    $\begingroup$ Dear Edgar, I cannot say how grateful I am for your reference to the question with the title: The distribution of the product of a Bernoulli & an exponential random variable. It is exactly the case that I mean. All my questions are answered there. Thanks again. $\endgroup$ Nov 16, 2019 at 13:40

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.