A C T A U N I V E R S I T A T I S L O D Z I E N S I S
F O L IA O E C O N O M IC A 194, 2005
ľadeusz Gerstenkorn*
A C O M P O U N D OF AN INFLATED PASCAL D ISTR IBU TIO N W ITH T H E P O ISS O N O N E
Abstract
In this p a p e r there is presented a c o m p o u n d o f an inflated Pascal d istrib u tio n with the P oisson one.
In the in tro d u c to ry p a rt o f the p a p er is giving an overview o f the last resu lts in topic o f co m p o u n d in g o f d istrib u tio n s, considering also the Polish resu lts. In succeeding Sections, p ro b ab ility function o f the com pound distrib u tio n Pascal-Poisson, factorial, crude and incom plete m o m en ts as well recu rren ce re la tio n s o f th is d istrib u tio n are presen ted .
M S C Iassilication: 60
Key words: c o m p o u n d d istrib u tio n s, c o m p o u n d inflated Pascal-P oisson d istrib u tio n , inflated Pascal d istrib u tio n , P oisson d istrib u tio n , facto rial, c rude and incom plete m o m en ts, recurrence re la tio n s for the m om ents.
I. IN T R O D U C T IO N
T h e problem o f the com pounding o f probability d istrib u tio n s goes back to the tw enties o f the 20lh century. It is w orth here to m en tio n the papers by M . G reenw ood and G .U . Y ule (1920; they to o k th e p aram eter o f the Poisson d istrib u tio n as a gam m a variate) and E.S. P earson (1925; he gave a device o f m aking the param eter o f a distributio n o f Bayes theorem ). In forties the problem was dealt w ith by L. L und berg (1940; the Pólya d istrib u tio n as a com pound one), F .E . S atterth w aite (see a rem ark on the com p ariso n o f S atterth w aite’s idea o f the generalized P oisson d istribu tion w ith a com p o u n d distribution in W. F eller’s paper, 1943; 390), W. Feller (1943; som e com pou nd d istributions as “ co n tag io u s” ones), G . Skellam (1948; the binom ial-beta giving Pólya-Eggenberger distribution), M .E. C ansado
* P ro fesso r, F a c u lty o f M ath em atics, U niversity o f Ł ó d ź and U niversity o f T ra d e , Ł ódź.
(1948; com poun d Poisson distributions). Som e in terrelatio n s am o ng com p o u n d and generalized d istribution s gave J. G urlan d (1957). Im p o rtan t theoretical problem s arc contained in papers by II. T cicher (I960, 1962), W. M o len aar (1965) and W. M o lenaar and W .R . Zw ct (1966). A lso in sixties we find the results by T. Ś ro d k a (1964; R ayleigh with gam m a, gam m a with R ayleigh; 1966 - Laplace with n o rm al N (0, 3) left trun cated at zero, with gam m a, R ayleigh, W cibull, exponential), by S.K. K a tti and J. G u rla n d (1967; Poisson-Pascal), D.S. D ubcy (1968; com p o u n d W cibull). T h en we have p apers by D. S. D ubcy (1970; co m p ou nd gam m a, beta and F ), T. G e rsten k o rn and Ś rodka (1972; generalized gam m a with gam m a), H . Jakuszcnkow (1973; Poisson with norm al N(m, n) left trun cated at zero, w ith generalized tw o -p a ram eter gam m a, with M axw ell an d R ayleigh), W. D yczka (1973; binom ial with beta (incom plete m om ents)) T. G ersten k orn (1982; binom ial with generalized beta), T. G e rsten k o rn (1994; gene ralized gam m a with exponential), T. G ersten k o rn (1996; Pólya with beta), G .E . W ilm ot (1989; Poisson-Pascal and others). M .L . H u an g and K .Y . b u n g (1993; D com pound Poisson d istributio n as a new extension o f the N eym an T ype A distribution).
A wide list of references to the problem considered on e can find in review studies by N .L. Joh n so n and S. K o tz (1969; ch a p te r 8: 183-215), G. P. Patil et al. (1968) and now adays by G. W im m er and G . A ltm ann (1999).
II. T H E C O M P O U N D IN G O F P R O B A B IL IT Y D IS T R IB U T IO N S
Below, we give a definition and relations needed for the com pounding o f d istribu tions.
Definition 1. T he com pound distribution
L et a ra n d o m variable X y have a d istributio n function ,Г(л:|у) depending on a p aram eter y. Suppose th a t the param eter у is considered as a random variable Y with a distribution function G(y). T h en the d istrib u tio n having the d istrib u tio n function o f X defined by
H (x ) =
J
F (x \c y)d G (y) (1)— 00
will be called co m pound, with th a t с is an arb itra ry co n stan t o r a con stan t bounded to som e interval (G urland, 1957; 265).
I he occurrcncc of the co n stan t с in (1) has a practical justification because the d istrib u tio n o f a rand om variable, describing a p henom enon,
often depends on the param eter which is a realization o f an o th er ran do m variable m ultiplied, how ever, by a certain constan t.
T h e variable which has distribution function (1) will be w ritten dow n in sym bol as “X л Y " and called a coi. .pound o f the variable X with respect to the “ co m p o u n d in g ” У. R elation (1) will be w ritten sym bolically as follows:
H (x ) = F (x \c y ) A G ( y ). (2)
У
C o n sid er the case w hen b o th variables are discrete, the first one with a probab ility fu nction P (X = X i \kn) depending on a p aram eter n being a ra n d o m variable N with a probability function P (N = n).
T h en (1) is expressed by the form ula
h(Xl) = P ( X = x t) = £ P ( X = x t\N = k n ) - P ( N = n). (3) /1=0
III. T H E IN F L A T E D PA S C A L л P O IS S O N Y
Definition 2. T h e distrib u tio n of the form
P£X = i\Y) = 1 — s + s ■ P ( X = j'ol ľ ) for i = i0,
s • P (X = i\Y ) for i = 0, 1, 2, i0 - l , i 0 + l , n,
(4)
where 0 < s < 1 and the variable У, being in the condition, has a distribution
G (y), we call an inflated conditional distribution. It is a discrete d istrib utio n
w ith an inflation o f the distrib u tio n P (X = i|Y ) at a point i = i0.
D istribu tion (4) were introduced, nam ed as inflated, by S.N. Singh (1963) and M .P . Singh (1965-66, 1966) and later discussed by m any au th o rs. Some in fo rm atio n a b o u t these results one can find in T. G e rsten k o rn (1977) and G. W im m er, G. A ltm an n (1999).
Lemma 1. T h e com pounding o f an inflated d istrib u tio n w ith an o th er d istrib u tio n gives a distrib u tio n with the same inflation.
T h e p ro o f, as ra th e r easy one, is here om itted.
Theorem 1. W e assum e th a t X is a v ariab le w ith in flated Pascal d istrib u tio n P a s{nk, p) depending on a p aram eter n, i.e.
Pas(nk, p) - Pa,( X = i\N = n) =
Cnkf°•"«
,0 * ---n i«! 1 — s + s (---p ° qnk for i = i01 №>•-». °
iS— j — p-qnk for i = 0, 1, 2, i0 ~ 1, i0 + 1, ... where p > 0, /c > 0, p + q = 1 and x 11' - 11 = jc(x + l)(x + 2)...(x + 1 - 1) is the so-called ascending factorial polynom e.Let N be a variable with the Poisson distrib u tio n Pa(n, Я)
P0( N = n, Я) = — e ~ A, n = 0, 1, 2, ..., Я > 0 .
n!
By using the form ula (3) we get
M O =
1 — s + s P - ; ‘olexp(0 — Я) for i = i0,
h'-for i = 0, 1, ..., i0 - 1, i0 + 1, ... (3)
w here p > О, Я > 0, 0 = qk and
Vm = Í * [i’ l]d F (x) (6)
is the so-called ascending (or reverse) factorial m o m en t determ ined by the equality:
(Г
Mm = Е[(кХУ ‘-~4 = X ( k n f ‘- exp( - 0),
n\
Proof. A t first we determ ine /1,(10) = P (X = i0): hs(i0) = P S(X = i0) = 00 rni-V'-- 1 ! J = £ ( l - s + s ----7 7— p ioqnk) - cxp( — Я) = n= 0 ■ 1 «о! п=о n\ (7)
A nalogously we calculate /1,(1').
If in (5) we take s = 1 we obtain the com pound Pascal-Poisson distribution w ith o u t inflation
h(i) = - - ^ ‘- е х р ( 0 - Л ) , / = 0, 1, 2, ...
R. Shum w ay and J. G urlan d (1960) have also o b tain ed this result but using a m ethod o f generalization o f distributions.
T h e m o m en t (6) or (7) is to be determ ined in dependence on crude o r factorial m om ent.
In the p ap e r by R. Shum w ay and J. G u rla n d (1960; 92) we find the follow ing form ula
м и = z s : k ‘ f s ‘j
/= i ;= i
w here Sj are Stirling num bers o f the second kind, к is a p aram eter and <*[,■) is a factorial m o m ent
“m = I x tadF (x) — 00
where x r‘] = x ( x — l)(x — 2 )...(x — i + 1) and S ' arc Stirling num bers o f the th ird kind.
In general the factorial m om ent a m are easier to be calculated th a n the crude ones. But one can pass from the factorial m o m ents to the crude ones.
T his problem has been exactly discussed in the p ap e r by T. G ersten k o rn (1983) where one can also find a table o f Stirling num b ers o f th e third kind.
IV . T H E C O M P O U N D P A S C A L -P O IS S O N D IS T R IB U T IO N A N D IT S M O M E N T S
T he com pound Pascal-Poisson w ithout inflation can be written symbolically
Pap(k, p, Я) ~ Pa(nk, p) л P 0(X)
П
T his d istrib u tio n has been presented in an equivalent form by S.K. K a tti and J. G u rla n d (1961) as a generalization o f the Poisson distribution w ith the P ascal one.
Corollary 1. T h e m o m en t o f the rlh order o f the co m p o u n d d istribution P ascal-P oisson is the m ean value o f an a p p ro p ria te m o m en t o f Pascal d istrib u tio n if the param eter n appearing in the form ulae o f m om ents o f the Pascal d istrib u tio n is considering as a ran d o m variable o f the Poisson d istrib u tio n , i. e.
fyr](k, P, X) = a[r](nk, p) л P 0( l )
П
br(k, p, A) = ar(nk, p) л P 0(A).
П
T ak in g in account these form ulae, we have
Theorem 2. T he factorial m om ent b[r](k, p, A) o f the com pound distribution P ascal-Poisson is expressed by
b[r](k, p, A) = u fá (8)
where ^ir] is the reverse factorial m om ent o f the Poisson d istrib u tio n , i.e.
t w !r' - n - , e x P( - X ) . (9)
n = 0 n '
Proof. b[r](k,p, A) = л В Д = ( ^ j ( ( n k ) [r~ 1] л Р Д ) ) =
A lready in G. B ohlm ann (1913; 398) we find the re latio n between the crude and factorial m om ents
br = Í s rjb lr] (10)
]= о
w here S j are Stirling num bers o f the second kind. T he table o f these n u m b e rs ca n be fo u n d , fo r in stan ce, in A. K a u fm a n n (1968) o r in J. Ł ukaszew icz and M. W arm us (1956).
Corollary 2. D irectly from (8) and (10) we have
br( k ,p , A) = W - Y / ť l (11)
j= o W
Rccurrcncc relations for crudc m oments
Theorem 3. C rude relations o f the com pound P ascal-P oisson Pap(k, p, A) d istrib u tio n arc expressed by
1) br+ t (k, p, X) = £ ( ) £ S}bu+ n(/c, p, A), (12) i = o V / J = 0 2) br+ |(fc, p, A) = Yj S fr v + n(k, P, A) -I- - ^-br(k, p, A). (13) 4 0 P ro of A d 1)
br+ i(k , p , A) = ar+ i(n/c, p) л P„(A) =
/I
= nkP„ í ( - V m + 1. p) a =
<11=0 V1/ "
= f Ź ( r\ n k a f r k + l , p ) A P„(А)). (14)
4 i-0 \ v я
Now we will calculate the expression in parentheses using (10), some properties o f factorial polynom ials, and the following proposition:
Proposition 1. F actorial m om en t o f the Pascal Pa(n k ,p ) d istrib u tio n is given by W ' ’ 11 (see: D yczka, 1973: 223 (63), (64)). n ka ^n k + 1, p) л P„(A) = П = I S lj ( - \ \ n k + l ) ^ - i ] л P 0(A) = 7=0 W = Ż S j^ Y (n fc(n fc + l))W --4 A P 0(A) = ;=o \ 9 / я = Z S j ^ W +1' - 1 , A P o(A )= j=o \ 9 / n ^=0
A d 2) W. D yczka (1973: 225 (70)) has show n th a t
a,+ i(n k, p) = P (nkar (nk + l , p ) + — / . ar(n k ,p )).
4
J p
A
Using this fo rm u la in (14) and regarding the first co m p o n en t, we have
nk ^ ar(nk + 1, p) л Pn{X) = nk ^ £ S ' ( J ) (nk + 1 ) [J-~ 11 л Pn( l ) = J
J
=
t ^ ^ Y +V ü b t , =is'jblj+
1](k,p,k).
J = 0 W j = o
I hen regarding the second co m ponent we have
P d ar(nk, p) л P0(X) =
J
d{ P df „/pV
P d u<ln
T h e result (13) is th en evident. Incomplete momentsA n incom plete (truncated, generalized) crude m om en t mr(t) o f o rd er r o f a d istrib u tio n function F (x) is defined by
00
mr(t) = jV d F (x ).
F o r incom plete factorial m om en t we have analogously mrn(0 = \ x [r]dF (x).
In the discrete case we assum e th a t t in the sum is an integer. It is quite evident th at a com plete m o m ent is a special case o f the incom plete one.
W. D y czka (1973: 212, (22)) has show n th a t incom plete cru de m om ents arc expressed by the incom plete factorial ones in th e follow ing way:
*=o
w here S', are Stirling num bers of the second kind. U sing this n o tio n for our case, we have
Theorem 4. Incom plete factorial m om ents o f the com pound Pascal-Poisson
Pap(k, p, A) d istrib u tio n are expressed by
where 0 = qkl and ß [r + 4 is the reverse factorial m o m en t given by (9). Proof. W . D yczka (1973: 227, (73)) has show n th a t
delink, p, t) = a[r](nk, p)a0(nk + r , p , t - r ) = i - о
C o m p o u n d in g this expression w ith P 0(A), we have
(<f*(n/c)tr+,,- 1]) л P 0(;.)) = A.
П
L et us calculate the expression in parantheses:
00 (ak) У
qnk(nkyr+i.-i] л p до = у (nfc)[,+1 .-1 1 H - ^ - e x p ( - Я) =
„=n n!
= exp(0 - A) £ (nk)[r+l- 11 exp( - 0) = exp(0 - ?)ц{г+п.
„ = r. n\
00 nn
T ak in g in account these calculations, we have A = t-r-lp T + l u w - E t1= n exp(° ~ ^ + = = br(k, p, A) - exp(0 - X)pr £ P.. ß [r+n-I = П »•
As a sim ple conclusion, we have
C orollary 3. Incom plete crude m om ents o f the co m p o u n d Pascal-Poisson
Pap(k, p, A) d istrib u tio n are expressed by
B o h lm an n G . (1913), F o rm u lieru n g und B egründung zw eier H ilfssätze d e r m athem atischen S tatistik , M ath. A nn., 74, 341-409.
C an sa d o M .E . (1948), O n the c o m p o u n d and generalized Poisson d istrib u tio n s, A nn. M ath. S la t., 19, 414-416.
D ubey D .S. (1968), Л co m p o u n d W eibull d istribution, N aval Res. le g is t. Q uart., 15, 2, 179-182. D u b ey D .S . (1970), C o m p o u n d g am m a, beta and F d istrib u tio n s, M e trik a , 16, 1, 27-31. D y czk a W. (1973), Z asto so w an ie sk lad an ia ro zk ład ó w d o w y znaczania m o m en tó w (A pplication
o f c o m p o u n d in g o f d istrib u tio n s to d e term in atio n o f m om ents; in Polish), Z e s z y ty N auk. Politech. łó d z k ie j, 168, M a tem aty k a 3, 205-230.
Feller W . (1943), O n general class o f „ co n tag io u s” d istrib u tio n s, A n n . M ath. S ta t., 14, 389-400. G ersten k o rn T ., Ś ro d k a T. (1972), Kom binatoryka i rachunek prawdopodobieństwa (C om binatorics
an d p ro b a b ility theory; in Polish), PW N , W arszaw a, V ll ed.: 1983.
G ersten k o rn T . (1977), Jednow ym iarow e rozkłady dyskretne ze zniekształceniem (Onc-dim cnsional inflated d iscrete d istrib u tio n s; in Polish), [in:] M eto d y sta tysty c zn e w sterow aniu jakością (S tatistical m eth o d s in q u a lity c o n tro l), ed. Sz. F irkow icz, R e p o rts on the conference o f the Polish A cad , o f Sciences (Ja b ło n n a, N o v . 23-28, 1975), O ssolineum , P olska A kadem ia N a u k , W rocław 1977, 195-208.
G e rsten k o rn T . (1982), T h e co m p o u n d in g o f the binom ial an d generalized beta distrib u tio n s, [in:] Proc. 2 '“‘ Pannonian Conf. on M athem . S ta tistics (T atzm a n n sd o rf, A u stria, 1981), ed. W. G ro ssm an n , G . C h. Pflug, W. W ertz, D. Reidel P ubl., D o rd re ch t, H o llan d , 87-99. G e rste n k o rn 'Г. (1983), R ela tio n s betw een the crude, facto rial and inverse facto ria l m om ents,
[in:] Trans, o f the N in th Prague Conf. on Inform ation Theory, S ta tistica l D ecision Functions and Random Processes (P rague, Ju n e 28 -Ju ly 2, 1982), A cad em ia P ubl. H ouse o f the C zechoslovak A cad, o f Sciences, P rague 247-252.
G e rsten k o rn T . (1994), A c o m p o u n d o f the generalized g am m a d is trib u tio n w ith the L aplace distribution truncated at zero, Bull. Soc. Sei. Letters Ix>dz, 44, Recherches sur les deformations,
hr( k ,p ,X ,t ) = X ( S ; / ) M( / c , p ,A ) - p j c x p ( f l - ; .) L
цИи+ц-r
R E F E R E N C E S
G e rsten k o rn T . (1996), Л c o m p o u n d o f the Pólya d is trib u tio n w ith the b e ta one, Random Oper, and Stoch. E qu., 4, 2, 103-110.
G reen w o o d M ., Y ule G .U . (1920), A n in q u iry in to the n a tu re o f freq u en cy d istrib u tio n rep resen tativ e o f m u ltip le h ap p en in g s with p a rticu la r reference to the occurance o f m ultiple a tta c k s o f disease or o f rep eated accidents, J. R r /. S ta t. Soc., 83, 255-279.
G urlan d J. (1957), Som e interrelations am ong com pound and generalized distributions, Biometrika, 44, 265-268.
H uang M .L ., F u n g K .Y . (1993), D com pound Poisson distribution, Statistical Papers, 34, 319-338. Jak u szen k o w H. (1973), N ow e złożenia ro z k ład ó w (N ew co m p o u n d s o f d istrib u tio n s; in Polish),
P rzegląd S ta ty sty c z n y , 20, 1, 67-73.
J o h n s o n N .L ., K o tz , S. (1969), D istributions in S ta tistics - D iscrete D istributions, H o u g h to n M ifflin C o m p ., B oston.
K a tti S .K ., G u rla n d J. (1961), T h e Poisson-Pascal d istrib u tio n , B iom etrics, 17, 527-538. K a u fm an n A . (1968), Introduction á la com binatorique en vue des applications, D u n o d , Paris. L u n d b erg O. (1940), On Random Processes and Their A pplication to Sickn ess a n d Accident
S tatistics, A lm q u ist and W iksells B oktryckeri - A. B., U ppsala.
Ł ukaszew icz J., W arm u s M . (1956), M e to d y num eryczne i graficzne (N u m erical an d graphical m eth o d s; in Polish), P W N , W arszaw a.
M o le n aa r W. (1965), Som e re m a rk s on m ixtures o f d istrib u tio n s, Bull. Intern. S ta tist. Institute, 35й1 Session o f the In tern . S tatist. In st., B eograd, B ook 2, 764-765.
M o ie n aa r W ., van Zw et W .R . (1966), O n m ixtures o f d istrib u tio n s, Ann. M a th . S ta t., 37, 1, 281-283.
Patii G .P ., Joshi S.W ., R ao C .R . (1968), A D ictionary and Bibliography o f Discrete Distributions, O liver and B oyd, E d in b u rg h .
Pearson E.S. (1925), Bayes’s theorem , exam ined in the light o f experim ental sam pling, Biometrika, 17, 388-442.
S a ttc rth w aile F .E . (1942), G eneralized Poisson d istrib u tio n , A nn. M ath. S ta t., 134, 410-417. Seweryn J.G . (1986), Som e probabilistic properties o f Bessel distributions, Z e sz yty N auk. Politechn.
Ł ó d z k ., 466, M a te m a ty k a 19, 69-87.
Shum w ay R ., G u rla n d J. (1960), A fitting p ro ced u re fo r som e generalized P oisson distrib u tio n s, Skandinavisk A ktu a rietid skrift, 43, 87-108.
Singh M .P . (1965-1966), Inflated binom ial d istrib u tio n , J. Sei. Res. Banares H indu Univ., 16, 1, 87-90.
Singh M .P . (1966), A n o te on generalized inflated binom ial d istrib u tio n , Ind. J. S la t., Ser. A , 28, 1, 99.
Singh S. N. (1963), A n o te o f inflated Poisson d istrib u tio n , J. Ind. S ta t. A ssoc., 1, 3, 140-144. S kellam J. G . (1948), A p ro b a b ility d istrib u tio n derived fro m th e binom ial d istrib u tio n by
re g ard in g the p ro b a b ility o f success as variable betw een th e sets o f trials, J . R oy. S tatist. S o c., Ser. B, 10, 2, 257-261.
Ś ro d k a T . (1964), N ouvelles com p o sitio n s de certaines d istrib u tio n s, Bull. Soc. Sei. In u r e s , 15, 5, 1-11.
Ś ro d k a T . (1966), Z łożenie ro z k ła d u L ap lace’a z pew nym uog ó ln io n y m ro zk ład em gam m a, M axw ella i W eibulla (C o m p o u n d in g o f the L aplace d istrib u tio n with a generalized gam m a, M axw ell a n d W cibull d istrib u tio n s; in P olish) Z e s z y ty N a u k . P olitech. Ł ó d z k., 77, W łókiennictw o, 14, 21-28.
T eicher H . (1960), O n th e m ixtures o f d istrib u tio n s, Ann. M ath. S ta t., 31, 55-72. T eichcr 11. (1962), Iden tifiab ility o f m ixtures, Ann. M ath. S ta t., 32, 244-248.
W illm ot G .E . (1989), A re m a rk o n the Poisson-P ascal and som e o th er c o n ta g io u s d istrib u tio n s, S ta tist. Probab. L ett., 1, 3, 217-220.
Tadeusz Gerstenkorn
ZŁOŻENIE INFLACYJNEGO ROZKŁADU PASCALA / . ROZKŁADEM POISSONA
Streszczenie
W p racy p rezen to w an e jest złożenie inflacyjnego ro zk ład u Pascala z ro zk ład em Poissona. W części w stępnej p racy p o d a n y je s t przegląd w yników badaw czych d otyczących tem atu złożeń ro zk ład ó w ze szczególnym uwzględnieniem polskich a u to ró w .
W d alszych rozd ziałach p o d a n o funkcję p raw d o p o d o b ień stw a ro zk ład u złożonego Pascal- -Poisson o ra z jeg o m o m en ty silniow e, zwykłe, niek o m p letn e o raz zw iązki rekurencyjne.