Zero-<strong>in</strong>‡ated modelsLambert [17] <strong>in</strong>troduced the Zero-In‡ated Poisson <strong>in</strong> order toac<strong>count</strong> for the excess number <strong>of</strong> zeros usually found <strong>in</strong> <strong>count</strong> regressions, which implyoverdispersion. It can be <strong>in</strong>terpreted as a mixture <strong>of</strong> the Poisson distribution f 2 () witha degenerate distribution whose mass is concentrated at zero. A b<strong>in</strong>ary process withdensity f 1 () sorts the outcomes between the two distributions. Thus, a zero can bean outcome either <strong>of</strong> the degenerate distribution or <strong>of</strong> the Poisson distribution, and theresult<strong>in</strong>g probabilities are:Pr [y i = 0] = f 1 (0) + (1 f 1 (0)) f 2 (0) (15)Pr [y i = r] = (1 f 1 (0)) f 2 (r) ; r = 1; 2::: (16)The b<strong>in</strong>ary distribution is ord<strong>in</strong>arily speci…ed as a logit with regressors z i and parameters.The log-likelihood then becomes:nXnXln L () = 1 (y i = 0) ln (exp (z 0 i) + exp ( exp (x 0 i)))+ (1 1 (y i = 0)) (y i x 0 i exp ( exp (x 0 i)i=1i=1The zero-<strong>in</strong>‡ated negative b<strong>in</strong>omial follows the same reason<strong>in</strong>g, just substitut<strong>in</strong>g thenegative b<strong>in</strong>omial for f 2 () .4.1.2 Ordered mult<strong>in</strong>omial modelsIn such models a …xed number <strong>of</strong> outcomes, say m, can be obta<strong>in</strong>ed, which can be thenordered <strong>in</strong>to <strong>in</strong>teger categories. Let the latent variable be:y i = x 0 i + u i (17)Then de…ne the observed outcomey i = j if j1 < y i jwhere the a j are the thresholds, or cuto¤ po<strong>in</strong>ts, to be estimated along with the parameters by MLE. It is conventioned that 0 = 1 and m = +1. Then16
Pr [y i = j] = Pr ( j 1 < yi j ) == F ( j x 0 i) F ( j 1 x 0 i) (18)where F () is the cdf <strong>of</strong> u i . An ordered logit is such model where the cdf is a logisticdistribution, and an ordered probit where the cdf is a Normal. In the former case, thereare K + m 1 parameters to be estimated, where K is the dimension <strong>of</strong> , while <strong>in</strong>the latter case, there are (m 1) (K + 1) parameters. These models have been applied to<strong>count</strong> data that take very few values. In some cases, as <strong>in</strong> our present study, the frequency<strong>of</strong> higher values is low enough so that an upper censor<strong>in</strong>g is undertaken (namely, entriesabove 5 are censored at that level).5 Data and resultsOur estimations are based on retail level market data. IMS Health is the ma<strong>in</strong> marketaudit company <strong>of</strong> the pharmaceutical <strong>in</strong>dustry worldwide. Pharmaceutical Market <strong>Brazil</strong>(PMB) surveys sales only through retail channel, so we are not able to <strong>in</strong>clude hospitalsales as expla<strong>in</strong><strong>in</strong>g variable. Only recently have hospital sales started be<strong>in</strong>g collected. Weuse then PMB data but exclude <strong>drug</strong>s sold typically or exclusively to hospitals, such asanesthesia, parenteral solutions, diagnostic agents, and blood derivatives. The data basehad also to be cleaned up with the exclusion <strong>of</strong> functional food, shampoos, soaps andother cosmetics. But phytotherapics have not been excluded.PMB records extracted comprised monthly sales <strong>in</strong> volume and value (local currency).Units are packages. Strength (amount <strong>of</strong> API content <strong>in</strong> the <strong>drug</strong>) is not available for all<strong>drug</strong>s <strong>in</strong> a separate …eld, but packag<strong>in</strong>g description is not standardized; as a consequence,the …eld for equivalent dosages is also <strong>in</strong>complete. We observe sales per brand and packag<strong>in</strong>g.Descriptive variables <strong>in</strong>clude: seller´s name; API; type <strong>of</strong> <strong>drug</strong> (reference, similaror <strong>generic</strong>) launch<strong>in</strong>g date <strong>of</strong> the packag<strong>in</strong>g; launch date <strong>of</strong> the brand or <strong>generic</strong>; formand route <strong>of</strong> adm<strong>in</strong>istration (e.g. tablets, sprays, ampoles, etc.). Follow<strong>in</strong>g Scott-Morton[24], [25] and advice from the <strong>in</strong>dustry 5 , we grouped forms <strong>in</strong>to (i) oral solids, (ii) oral5 We are extremely grateful to Mr. Jair Calixto, manager <strong>of</strong> good manufactur<strong>in</strong>g practices and phar-17