In most cases, analysis of insect counts data has been modelled by the three distributions, namely Poisson (θ), Binomial (n, θ) and Negative Binomial (k, θ). These discrete distributions fall under the category of power series distributions. These distributions have been generalized to what is referred to as the generalized power series distribution (GPSD). By expressing these probability distributions explicitly in the form of power series distributions, this has greatly simplified the derivation of the explicit form of the moments of these distributions. The power series distributions have been widely studied. Noak  specifically investigated its moments and cumulants properties which were later extended by Khatri  to multivariate distributions. The properties of the generalized power series distributions were later studied by Patil  who also investigated estimation and other properties of these models. In more recent studies, a wider interest has been focused on modifications of the power series probability distributions to address the problem of zero-inflation, i.e. observing more zeroes than what would be expected for example, in a Poisson or Negative binomial distribution. Count data has been noted to consist of more zeroes than expected if the data were generated by a simple Poisson or negative Binomial process. Ignoring zero-inflation can have two consequences: bias may be induced in the estimation of model parameters and standard errors may be inflated thus leading to erroneous statistical inferences. In addition, zero-inflated data may as a consequence, cause over-dispersion which is defined as having a variance that may be larger than the mean. If this data characteristic is left unaddressed, this will render the model specification for the Poisson distribution of equal mean and variance null and void. Cameron and Trivedi  have outlined some common departures from the standard Poisson regression that should be addressed before actual analysis. Dean  outlined a method of testing for over-dispersion in Poisson and Binomial regression models. Lately however, the problem of zero inflation has been widely researched in the context of zero-inflated count regression. See for example, Lambert  and Gupta, et al.   . Most of the cited research however, has implemented count regression models.
2. Power Series Distributions
A discrete random variable X will have a power series distribution given as:
The distribution belongs to the exponential family of distributions and can generally be expressed in the form:
where a and g are functions of the unknown parameter θ and c is a function of x.
This property has been exploited in the derivation of the moments and other properties of the distribution. It can be shown (see Edwin  ) that the central moments for the power distribution are given as:
Table 1 gives a summary of some power series distributions represented in the form given in (1).
3. Zero-Inflation and Over-Dispersion
The Poisson distribution theoretically specifies that the mean and variance are equal. However, it is quite common to have data for which the variance is far larger than the mean and the phenomenon is referred to as over-dispersion. In this case, Poisson GLM has been used to correct the anomaly. See for example, Cameron and Trivedi  . Over-dispersion is also common in a Negative Binomial distribution. When the distribution is parameterized in terms of its mean μ and variance σ2 distribution, the mean and variance are given as;
The distribution is reduced to equi-distribution ask becomes large, implying convergence to the Poisson distribution. As k becomes small for a small μ, a zero-inflated Negative Binomial distribution is a consequence.
A zero-inflated statistical model is based on a zero-inflated probability distribution. It arises when probability mass at a point zero exceeds the one allowed under the particular family of distributions. These models have been widely studied. See for example, Jasankul  , Ridout et al.  and Bohning  . They provide one method to explain the excess zeroes by modelling the data as a mixture of two separate distributions in which one is typically a Poisson or Negative Binomial distribution that can generate both zeroes and non-zero counts, and the second distribution is a constant distribution that generates only zero counts.
4. The Zero-Inflated Distribution
Count data that have an incidence of zeroes greater than expected for the underlying probability distribution is modelled as:
Table 1. Specification of some common power series distributions.
where is the zero-inflation parameter. in this case represents any one of the count data distributions, e.g. Poisson, Negative Binomial, etc. using the format presented in (1). The means and variances of these distributions may then be obtained by using the expressions in (3) for the respective probability model. For example, for the Poisson (λ) distribution;
And for a Negative Binomial distribution;
Using (5) and (1) and applying the usual definitions for E(X) and V(X), the mean and variance of a zero-inflated distribution is given as:
where is given in Table 1 for the respective probability distribution.
5. Estimation of Model Parameters
In estimating the zero inflated parameters, some three methods seem to have been prominently used. The methods of moments (MM) is said to provide estimates which are not very accurate. Nanjundan and Naika  ,  applied the MM as well as the maximum likelihood (ML) estimation and concluded that both methods are asymptotically equally efficient. In some cases however, the MM is considered as appropriate for use as initial values for ML estimation in the case of a non-closed solution of the model parameters. The two methods are outlined below:
5.1. Method of Moments Estimation
Let be a random sample from the probability distribution of the form in (5) for a p-parameter ZI model, the estimates are obtained by solving the equations:
For a two-parameter ZI model, Equation (9) leads to the following two equations:
Equation (10) may then be solved simultaneously to obtain estimates for ρ and θ.
5.2. Maximum Likelihood Estimation
The likelihood function for a ZI model given in Equation (5) may be written as:
The log-likelihood function is then given as:
where n0 is the number of zeros in the observed sample. Solving and for θ and ρ, the following expressions are obtained after some simplification:
The Newton-Raphson iterative method usually provide a solution of the form . For an initial value of θn, the next estimate is given as
Applying the Newton-Raphson iterative method to Equation (14) and using the MM estimates from Equation (10) as initial values then the improved estimate of θ is:
Finally, substituting the estimate from Equation (15) into Equation (13), an estimate of is obtained. We shall now apply the estimation procedures outlined above to two probability distributions, namely ZIP and ZINB as given in Equation (6) and Equation (7) respectively.
5.3. Fitting the ZI-Poisson Model
For the ZI-Poisson distribution, the corresponding function f(θ) from Table 1 is substituted in Equation (10) and solved simultaneously to obtain the closed form MM estimates for ρ and θ as:
Likewise, the ML estimates for ZIP model are obtained by substituting f(θ) in Equation (13)and Equation (14) resulting into:
Using the MM estimate for θ as the initial value, Equation (18) is solved iteratively and the estimate is subsequently substituted in Equation (19) to obtain .
5.4. Fitting the ZI-Negative Binomial Model
To analyse the data using this model, we need to obtain a pooled estimate of k. Anscombe  has indicated that fitting a negative binomial distribution with a common value of k to sets of counts on the same species is a reasonable procedure. He then proposed such an estimate to be obtained from Equation (4) by substituting μ and σ2 with the sample values and s2 and solve for k. For the count data the estimate of k is 0.12.
Again the corresponding for the Negative binomial distribution from Table 1 is substituted in Equation (10) and solved simultaneously for and to obtain the MM estimates as:
Similarly, for the ML estimates, we substitute the respective from Table 1 in (13) and solve to obtain the following estimates:
Equation (22) is solved iteratively for and substituted in (23) to obtain .
The data to be used are some counts of eggs of Aphis fabae made by Dr. D. P. Jones in the course of a survey of the Eastern counties of England in 1947 and reproduced by Anscombe  . In this paper, the data has been sorted into a slightly different frequency distribution form but the original data remains the same. Ninety four hedgerow spindle sites were visited, that had been cut down the previous winter, so that the shoots were of one year growth. At each site ten shoots were removed and the A. fabae eggs on them subsequently counted. The counts are given in Table 2.
The mean and variance of the egg counts is 5.29 and 235.01 respectively which points to a strong evidence of over-dispersion. A frequency distribution for the counts is given in Figure 1. The large spike at zero gives early warning of possible zero inflation. Therefore, the ZIP and ZINB distributions are to be fitted.
7. Results and Conclusions
Table 3 gives the estimated parameters for the count data on the basis of fitting
Table 2. Egg counts.
Table 3. Parameter estimates.
Figure 1. Frequency distribution.
the ZIP and ZINB distributions. It is noted that the estimated value of the exponent for ZINB model from the sample version of Equation (4) was k = 0.12 which in turn, resulted in a negative estimate of the inflation parameter ρ, that may be interpreted as a case of zero-deflation. This may have been a consequence of the observed heterogeneity between sites which was not addressed here and which ranged between 0 and 2032 within sites. As proposed by Anscombe  , for a k < 1 a logarithmic transformation of the counts may be necessary. This problem may be addressed in a future analysis. However, the situation may not arise when count regression model is used.
It should be noted however, that possible design errors, such as sampling practices, might also have caused the excess zeros.
Below are charts representing the Poisson frequency distribution. Figure 1 represents the observed frequencies of the count data which is highly positively skewed and Figure 2 represents the Poisson expected frequencies which is centred at around five eggs. That means if we assume a Poisson distribution for this data, then we would expect approximately less than 10 zeros. Figure 3 gives a plot of the ZIP fitted frequencies. The zero-inflation factor does not differ much from the original counts but overall, the counts are centred at five eggs. It is
Figure 2. Expected frequencies.
Figure 3. ZIP frequencies.
therefore recommended that before implementing procedures to address the zero-inflation, the inherent heterogeneity between sites be further examined.
 Dean, C.B. (1992) Testing for Over Dispersion in Poisson and Binomial Regression Models. Journal of the American Statistical Association, 87, 451-457.
 Gupta, P.L. and Gupta, R.C. (1995) Inflated Modified Power Series Distribution with Applications. Communication in Statistics-Theory and Methods, 24, 2355-2374.
 Gupta, P.L., Gupta, R.L. and Tripathi, R.C. (2004) Score Test for Zero Inflated Generalized Poisson Regression Model. Communication in Statistics-Theory and Methods, 33, 47-64.
 Ridout, M., Hinde, J. and Demetrio, C.G.B. (2001) A Score Test for Testing Zero-Inflated Poisson Regression Model against Zero Inflated Negative Binomial Alternatives. Biometrics, 57, 219-223.
 Bohning, D. (1998) Zero-Inflated Poisson Models and CAMAN. A Tutorial of Evidence. Biometrical Journal, 40, 833-843.
 Nanjundan, G. and Naika, T.R. (2012) Asymptotic Comparison of Method of Moments Estimators and Maximum Likelihood Estimators of Parameters in Zero-Inflated Poisson Model. Applied Mathematics, 3, 610-616.