Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf ·...
Transcript of Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf ·...
![Page 1: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/1.jpg)
Leslie Matrix II
Formal DemographyStanford Spring Workshop in Formal Demography
May 2008
James Holland JonesDepartment of Anthropology
Stanford University
May 2, 2008
1
![Page 2: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/2.jpg)
Outline
1. Spectral Decomposition
2. Keyfitz Momentum
3. Transient Dynamics in terms of Eigenvectors
4. Matrix Perturbations
(a) Sensitivity(b) Elasticity(c) Second Derivatives
Stanford Summer Short Course: Leslie Matrix II 2
![Page 3: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/3.jpg)
Spectral Decomposition of the Projection Matrix
Suppose we are given an initial population vector, n(0)
We can write n(0) as a linear combination of of the right eigenvectors, ui of theprojection matrix A
n(0) = c1u1 + c2u2 + · · ·+ ckuk
where the ci are a set of coefficients
We can collect the eigenvectors into a matrix and the coefficients into a vector andre-write this equation as:
n(0) = Uc
Stanford Summer Short Course: Leslie Matrix II 3
![Page 4: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/4.jpg)
From this it is clear that
c = U−1n(0)
Now, U−1 is just the matrix of left eigenvectors (or their complex conjugatetranspose), so:
ci = v∗i n(0)
Stanford Summer Short Course: Leslie Matrix II 4
![Page 5: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/5.jpg)
Spectral Decomposition II
Project the initial population vector n(0) forward by multiplying it by the projectionmatrix A
n(1) = An(0)
=∑
i
ciAui
=∑
i
ciλiui
Multiply again!
n(2) = An(1)
Stanford Summer Short Course: Leslie Matrix II 5
![Page 6: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/6.jpg)
=∑
i
ciλiAui
=∑
i
ciλ2iui
We could keep going, but at this point it isn’t hard to believe that the followingholds:
n(t) =∑
i
ciλtiui (1)
Equivalently:
n(t) =∑
i
λtiuiv∗i n(0) (2)
This is known as the Spectral Decomposition of the projection matrix A
It is instructive to compare this to the solution for population growth in anunstructured (i.e., scalar) population, characterized by a geometric rate of increasea:
Stanford Summer Short Course: Leslie Matrix II 6
![Page 7: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/7.jpg)
N(t + 1) = aN(t)
N(t) = N(0)at
For the scalar case, the solution is exponential
For a k-dimensional matrix, this solution means that the population size at time tis a weighted sum of k exponentials
While both depend on the initial conditions, the k-dimensional case weights theinitial population vector by the reproductive values of the k classes
Stanford Summer Short Course: Leslie Matrix II 7
![Page 8: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/8.jpg)
Population Momentum
0.00 0.05 0.10 0.15
020
4060
80
Fraction of Stable Population
Age
growingstationary
Stanford Summer Short Course: Leslie Matrix II 8
![Page 9: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/9.jpg)
Keyfitz (1971) Formulation for Population Momentum
The ratio of population size at the ultimate equilibrium and just before the fertilitytransition
M =b◦e0
rµ
(R0 − 1
R0
)(3)
where b is the gross birth rate:
b =1∫ β
αe−rxl(x)dx
◦e0 is the life expectancy at birth
Stanford Summer Short Course: Leslie Matrix II 9
![Page 10: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/10.jpg)
µ is the mean age of childbearing in the stationary population:
µ =
∫ β
αxl(x)m(x)dx∫ β
αl(x)m(x)dx
R0 is the net reproduction ratio:
R0 =∫ β
α
l(x)m(x)dx
Stanford Summer Short Course: Leslie Matrix II 10
![Page 11: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/11.jpg)
Momentum in Terms of Matrix Formalism
First recall the spectral decomposition of the projection matrix A:
n(t) =∑
i
ciλtiui
where the coefficients of the sum ci are
ci = v∗i n(0)
The (scalar) product of the initial population size and the reproductive value vectorcorresponding to the ith eigenvalue
Stanford Summer Short Course: Leslie Matrix II 11
![Page 12: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/12.jpg)
Momentum is the Ratio of Eventual Size to the Size atFertility Drop
M = limt→∞
‖n(t)‖‖n(0)‖
(4)
where ‖n‖ =∑
i ni is the total population size
The population under the old and new rates will be characterized by eigenvalues,
λ(old)i and λ
(new)i
limt→∞
n(t)
λ(new)1
= limt→∞
n(t) (5)
=(v(new)∗
1 n(0))u(new)
1 (6)
Stanford Summer Short Course: Leslie Matrix II 12
![Page 13: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/13.jpg)
This follows directly from the spectral decomposition of the projection matrix
Substitute this into equation 4
M =eT
(v(new)∗
1 n(0))u(new)∗
eTn(0)(7)
where e is a vector of ones.
This much simpler computationally than the original Keyfitz (1971) formulation
Stanford Summer Short Course: Leslie Matrix II 13
![Page 14: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/14.jpg)
Transient Dynamics and Convergence
Re-write the spectral decomposition for the projection matrix A, expanding the sumfor expository purposes
n(t) = c1λt1u+c2λ
t2u2 + · · ·+ ckλ
tkuk
If A is irreducible and primitive, then the Perron-Frobenius theorem insures thatone of the eigenvalues, λ1, of A will be:
• Strictly positive
• Real
• Greater than or equal to all other eigenvalues
Because of this, all the exponential terms in the spectral decomposition of A becomenegligible compared to the one associated with λ1 when t gets large
To see this divide both sides of the spectral decomposition of A by the dominanteigenvalue raised to the t power λt
1:
Stanford Summer Short Course: Leslie Matrix II 14
![Page 15: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/15.jpg)
n(t)λt
1
= c1u1 + c2
(λ2
λ1
)t
u2 + · · ·+ ck
(λk
λ1
)t
uk (8)
Since λ1 > λi for i ≥ 2, each of the fractions involving the eigenvalues will approachzero as t →∞
Taking this limit, we see that:
limn→∞
n(t)λt
1
= c1u1 (9)
This is the Strong Ergodic Theorem
The long term dynamics of a population governed by primitive matrix A aredescribed by growth rate λ1 and population structure u1
Stanford Summer Short Course: Leslie Matrix II 15
![Page 16: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/16.jpg)
Yes, But How Soon to Convergence?
Clearly, a population will converge to its asymptotic behavior faster is λ1 is largerelative to the other eigenvalues
That’s because λi/λ1 (i > 1) will be smaller and approach zero more rapidly as itis powered up
The classic measure of convergence is the ratio of the dominant to the absolutevalue of the subdominant eigenvalue
ρ = λ1/|λ2|
The quantity ρ is known as the damping ratio
From equation 8, we can write
limt→∞
(n(t)λ1
− c1u1 + c2ρ−tu2
)= 0
Stanford Summer Short Course: Leslie Matrix II 16
![Page 17: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/17.jpg)
For large t and some constant k
∥∥∥∥n(t)λ1
− c1u1
∥∥∥∥ ≤ kρt
= ke− log ρ
Convergence to stable structure (and therefore growth at rate λ1) is asymptoticallyexponential at a rate at least as fast as log ρ
Note that convergence could be faster (e.g., if n(0) = u1)
The time tx it takes for the influence of λ1 to be x times larger than λ2 is
(λ1
|λ2|
)tx
= x
or
Stanford Summer Short Course: Leslie Matrix II 17
![Page 18: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/18.jpg)
tx = log(x)/ log(ρ)
Stanford Summer Short Course: Leslie Matrix II 18
![Page 19: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/19.jpg)
How Far is a Population from the Stable Age Distribution?
The classic measure of distance is attributable to Keyfitz (1968) and is a standardmeasure of the distance between two probability vectors
∆(x,u) =12
∑i
|xi − wi|
Keyfitz’s ∆ takes a maximum value of ∆ = 1, clearly, ∆ = 0 when two vectors areidentical
Cohen developed two metrics that account not just for the population vector but forthe possible route through which a population structure can converge to stability
s(A,n(0), t) =t∑
u=0
(n(u)λu
1
− c1u1
)Stanford Summer Short Course: Leslie Matrix II 19
![Page 20: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/20.jpg)
r(A,n(0), t) =t∑
u=0
∣∣∣∣n(u)λu
1
− c1u1
∣∣∣∣Cohen’s measures of distance between the observed population vector and the stablevector is simply the sum of the absolute values of these vectors as t →∞
D1 =∑
i
limt→∞
|si(A,n(0), t)|
D1 =∑
i
limt→∞
|ri(A,n(0), t)|
Stanford Summer Short Course: Leslie Matrix II 20
![Page 21: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/21.jpg)
Matrix Perturbations
This derivation follows Caswell (2001)
We start from the general matrix population model:
Aw = λu (10)
Now we perturb the system
(A + dA)(u + du) = (λ + dλ)(u + du) (11)
Multiply all the products and discard the second-order terms such as (dA)(du)
Aw + A(du) + (dA)u = λu + λ(du) + (dλ)u (12)
Simplify this to yield
Stanford Summer Short Course: Leslie Matrix II 21
![Page 22: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/22.jpg)
A(du) + (dA)u = λ(du) + (dλ)u (13)
Multiply both sides by v∗ to get
v∗A(du) + v∗(dA)u = λv∗(du) + v∗(dλ)u (14)
From the definition of a left eigenvector, we know that the first term on the left-handside is the same as the first term on the right-hand side. Similarly, because the rightand left eigenvectors are scaled so that 〈u,v〉 = 1, the last term simplifies to dλ.We are left with
v∗dAu = dλ (15)
When we do a perturbation analysis, we typically only change a single element ofA. Thus the basic formula for the sensitivity of the dominant eigenvalue to a smallchange in element aij is
∂λ
∂aij= viuj (16)
Stanford Summer Short Course: Leslie Matrix II 22
![Page 23: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/23.jpg)
In other words, the sensitivity of fitness to a small change in projection matrixelement aij is simply the ith element of the left eigenvector weighted by theproportion of the stable population in the jth class (assuming vectors have beennormed such that 〈vu〉 = 1)
Stanford Summer Short Course: Leslie Matrix II 23
![Page 24: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/24.jpg)
Eigenvalue Sensitivities are Linear Estimates of the Change inλ1, Given a Perturbation
λ
ija
Stanford Summer Short Course: Leslie Matrix II 24
![Page 25: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/25.jpg)
Lande’s Theorem
Imagine a multivariate phenotype (e.g., a life history) z
Lande (1982) shows that when selection is weak, the change in the phenotype ∆zis given by
∆z = λ−1 Gs
wher G is the additive genetic covariance matrix for the transitions in the life-cycleand s is a column vector of all the life-cycle transition sensitivities
The sensitivities therefore represent the force of selection on the phenotype
Stanford Summer Short Course: Leslie Matrix II 25
![Page 26: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/26.jpg)
Some R Code for Calculating Sensitivities
A recipe!
> lambda <- eigen(A)> U <- lambda$vectors> u <- abs(Re(U[,1]))> V <- solve(Conj(U))> v <- abs(Re(V[1,]))> s <- v%o%w> s[A == 0] <- 0
Extract and plot them
> surv.sens <- s[row(s) == col(s)+1]> fert.sens <- s[1,]> age <- seq(0,45,by=5)> plot(age,surv.sens,type="l",lwd=3,col="turquoise")> lines(age[4:10], fs[4:10], lwd=3, col="magenta")> legend(30,.17,c("survival","fertility"), lwd=3, col=c("turquoise","magenta"))
Stanford Summer Short Course: Leslie Matrix II 26
![Page 27: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/27.jpg)
0 10 20 30 40
0.00
0.05
0.10
0.15
Age
Sen
sitiv
itysurvivalfertility
Stanford Summer Short Course: Leslie Matrix II 27
![Page 28: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/28.jpg)
Elasticities
Another measure of the change in a matrix given a small change in an underlyingelement is the eigenvalue elasticity:
eij =∂ log λ
∂ log aij
Elasticities are proportional sensitivities: they measure the linear change on a logscale
An important property of elasticities is that they sum to one
∑i,j
eij = 1
Stanford Summer Short Course: Leslie Matrix II 28
![Page 29: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/29.jpg)
0 10 20 30 40
0.00
0.05
0.10
0.15
Age
Ela
stic
itysurvivalfertility
Stanford Summer Short Course: Leslie Matrix II 29
![Page 30: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/30.jpg)
The Universality of the Human Life Cycle
There is a great variety of human demographic experience
How does this variation translate into the force of selection on the human life cycle?
Sampling the full variety of human demographic variation is hopeless
Adopt a strategy (following Livi-Bacci) of filling out a demographic space andexploring the boundaries of the region
Stanford Summer Short Course: Leslie Matrix II 30
![Page 31: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/31.jpg)
The Human Demographic Space
2 3 4 5 6 7 8
20
30
40
50
60
70
80
Total Fertility Rate
e0
−0.02
−0.020
0
0
0
00.02
0.02
0.02
0.04
Stanford Summer Short Course: Leslie Matrix II 31
![Page 32: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/32.jpg)
Mortality
0 20 40 60 80 100
−8
−6
−4
−2
Age
Log(
Mor
talit
y R
ate)
!KungAcheVenUSA
Stanford Summer Short Course: Leslie Matrix II 32
![Page 33: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/33.jpg)
Fertility
0 10 20 30 40 50
0.00
0.05
0.10
0.15
0.20
0.25
0.30
Age
AS
FR
!KungAcheVenUSA
Stanford Summer Short Course: Leslie Matrix II 33
![Page 34: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/34.jpg)
Survival Elasticities
0 10 20 30 40
0.00
0.05
0.10
0.15
0.20
Age
Ela
stic
ity!KungAcheVenUSA
Stanford Summer Short Course: Leslie Matrix II 34
![Page 35: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/35.jpg)
Fertility Elasticities
0 10 20 30 40 50
0.00
0.05
0.10
0.15
0.20
Age
Ela
stic
ity!KungAcheVenUSA
Stanford Summer Short Course: Leslie Matrix II 35
![Page 36: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/36.jpg)
Eigenvalue Second Derivatives
We can measure the curvature of the fitness surface
∂2λ(1)
∂aij∂akl= s
(1)il
∑m6=1
s(m)kj
λ(1) − λ(m)+ s
(1)kj
∑m6=1
s(m)il
λ(1) − λ(m), (17)
where m is the rank of the projection matrix,
s(m)ij = ∂λ(m)/∂aij,
and λ(m) is the mth eigenvalue of the projection matrix
Stanford Summer Short Course: Leslie Matrix II 36
![Page 37: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/37.jpg)
Sensitivity of Elasticities
One application of the eigenvalue second derivatives is calculating how elasticitieswill change when vital rates are perturbed
i.e., the sensitivities of the elasticities:
∂eij
∂akl=
aij
λ
∂2λ
∂aij∂akl− aij
λ2
∂λ
∂aij
∂λ
∂akl+
δikδjl
λ
∂λ
∂aij(18)
where δik and δjl indicate the Kronecker delta function where δik = 1 if i = k,otherwise δik = 0
Stanford Summer Short Course: Leslie Matrix II 37
![Page 38: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/38.jpg)
Sensitivities of Fertility Elasticities (Fertility)
10 20 30 40
0.00
0.05
0.10
0.15
Age (Fertility)
∂∂2 λλ∂∂F
i2
F10
10 20 30 40
0.00
0.05
0.10
0.15
Age (Fertility)
∂∂2 λλ∂∂F
i2
F15
10 20 30 40
−0.
050.
000.
050.
100.
15
Age (Fertility)
∂∂2 λλ∂∂F
i2
F20
10 20 30 40
−0.
050.
000.
050.
10
Age (Fertility)
∂∂2 λλ∂∂F
i2
F25
10 20 30 40
−0.
050.
000.
050.
10
Age (Fertility)
∂∂2 λλ∂∂F
i2
F30
10 20 30 40
0.00
0.05
0.10
Age (Fertility)
∂∂2 λλ∂∂F
i2
F35
10 20 30 40
0.00
0.04
0.08
0.12
Age (Fertility)
∂∂2 λλ∂∂F
i2
F40
10 20 30 40
0.00
0.04
0.08
0.12
Age (Fertility)
∂∂2 λλ∂∂F
i2
F45
10 20 30 40
0.00
0.04
0.08
0.12
Age (Fertility)
∂∂2 λλ∂∂F
i2
F50
Stanford Summer Short Course: Leslie Matrix II 38
![Page 39: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/39.jpg)
Sensitivities of Survival Elasticities (Survival)
0 5 10 15 20 25 30 35
−0.
010.
000.
010.
02
Age (Survival)
∂∂2 λλ∂∂S
i2
S0
0 5 10 15 20 25 30 35
−0.
020.
000.
01
Age (Survival)
∂∂2 λλ∂∂S
i2
S5
0 5 10 15 20 25 30 35
−0.
020.
000.
010.
02
Age (Survival)
∂∂2 λλ∂∂S
i2
S10
0 5 10 15 20 25 30 35
−0.
010.
010.
03
Age (Survival)
∂∂2 λλ∂∂S
i2
S15
0 5 10 15 20 25 30 35
−0.
010.
010.
03
Age (Survival)
∂∂2 λλ∂∂S
i2
S20
0 5 10 15 20 25 30 35
−0.
010.
010.
03
Age (Survival)
∂∂2 λλ∂∂S
i2
S25
0 5 10 15 20 25 30 35
−0.
010.
000.
010.
02
Age (Survival)
∂∂2 λλ∂∂S
i2
S30
0 5 10 15 20 25 30 35
−0.
005
0.00
50.
015
Age (Survival)
∂∂2 λλ∂∂S
i2
S35
0 10 20 30 40
0.00
00.
010
0.02
0Age
∂∂2 λλ∂∂S
i2
S40
Stanford Summer Short Course: Leslie Matrix II 39
![Page 40: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/40.jpg)
Sensitivities of Survival Elasticities (Fertility)
0 10 20 30 40
−0.
100.
000.
10
Age (Fertility)
∂∂2 λλ∂∂S
iFj
S0
0 10 20 30 40
−0.
100.
000.
05
Age (Fertility)
∂∂2 λλ∂∂S
iFj
S5
0 10 20 30 40
−0.
050.
000.
05
Age (Fertility)
∂∂2 λλ∂∂S
iFj
S10
0 10 20 30 40
−0.
08−
0.04
0.00
0.04
Age (Fertility)
∂∂2 λλ∂∂S
iFj
S15
0 10 20 30 40
−0.
06−
0.02
0.02
0.06
Age (Fertility)
∂∂2 λλ∂∂S
iFj
S20
0 10 20 30 40
−0.
040.
000.
040.
08
Age (Fertility)
∂∂2 λλ∂∂S
iFj
S25
0 10 20 30 40
−0.
020.
020.
060.
10
Age (Fertility)
∂∂2 λλ∂∂S
iFj
S30
0 10 20 30 40
0.00
0.04
0.08
0.12
Age (Fertility)
∂∂2 λλ∂∂S
iFj
S35
0 10 20 30 40
0.00
0.04
0.08
0.12
Age (Fertility)
∂∂2 λλ∂∂S
iFj
S40
Stanford Summer Short Course: Leslie Matrix II 40
![Page 41: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/41.jpg)
Sensitivities of Fertility Elasticities (Survival)
0 10 20 30 40
−0.
005
0.00
50.
010
Age (Survival)
∂∂2 λλ∂∂F
iSj
F10
0 10 20 30 40
−0.
020.
00
Age (Survival)
∂∂2 λλ∂∂F
iSj
F15
0 10 20 30 40
−0.
020.
00
Age (Survival)
∂∂2 λλ∂∂F
iSj
F20
0 10 20 30 40
−0.
015
0.00
00.
010
0.02
0
Age (Survival)
∂∂2 λλ∂∂F
iSj
F25
0 10 20 30 40
−0.
005
0.00
50.
015
Age (Survival)
∂∂2 λλ∂∂F
iSj
F30
0 10 20 30 40
−0.
005
0.00
5
Age (Survival)
∂∂2 λλ∂∂F
iSj
F35
0 10 20 30 40
−0.
004
0.00
20.
006
Age (Survival)
∂∂2 λλ∂∂F
iSj
F40
0 10 20 30 40
−5e
−04
5e−
04
Age (Survival)
∂∂2 λλ∂∂F
iSj
F45
Stanford Summer Short Course: Leslie Matrix II 41
![Page 42: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/42.jpg)
More Uses for Sensitivities: Stochastic Population Growth
Assume a population living in an i.i.d. random environment
The population is characterized by a mean projection matrix A, with its associatedeigenvalue λ
Tuljapurkar’s small noise approximation for the stochastic growth rate is:
log λs = log λ− σ2
2
where
σ2 = sT Cs
where C is the covariance matrix for life-cycle transitions across time
Stanford Summer Short Course: Leslie Matrix II 42
![Page 43: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/43.jpg)
When the environment is not i.i.d., there is a second term in the formula for log λs
to account for the autocorrelation
This term also contains sensitivities – the transient effects of autocorrelation will begreater for transitions to which λ is more sensitive (see Tuljapurkar & Haridas 2006)
Stanford Summer Short Course: Leslie Matrix II 43
![Page 44: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/44.jpg)
Migration Matrices
Imagine a population divided between two regions where there is migration betweenthe regions
Rogers (1966) introduced the multi-regional model, which has the following structure
Multi-regional models take the form of block matrices
The matrices along the digonal of the super-matrix are the Leslie matrices trackingsurvival and fertility within the subpopulations
The matrices on the off-diagonals of the super-matrix track the movement betweenthe different sub-populations
R =(
A1 M1←2
M2←1 A2
)(19)
A1 is the Leslie matrix for the population in region 1, A2 is the Leslie matrix forthe population in region 2
Stanford Summer Short Course: Leslie Matrix II 44
![Page 45: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/45.jpg)
M1←2 is the matrix moving individuals from region 2 to region 1 and M2←1 is thematrix moving individuals from region 1 to region 2
Stanford Summer Short Course: Leslie Matrix II 45
![Page 46: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/46.jpg)
Migration Matrices 2
We can draw a life-cycle graph for this simple multi-regional model
3’
1P2
F2F3
M1M2
M’2M’1
P’1 P’2
F’2 F’3
1 2 3
1’ 2’
P
Stanford Summer Short Course: Leslie Matrix II 46
![Page 47: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/47.jpg)
Migration Matrices 3
Assume there are three age classes, with reproduction in the last two
Migration is bidirectional and we assume that a migrant of age i at time t becomesage i + 1 at t + 1
The multi-regional matrix R accounts for the survival and reproduction of the twosub-populations as well as the movement between them
Denote the survival probabilities and fertility and migration rates of the firstpopulation: Pi, Fi, and Mi
Denote the same rates for the second population as P ′i , F ′i , and M ′i
Stanford Summer Short Course: Leslie Matrix II 47
![Page 48: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/48.jpg)
Migration Matrices 4
The multi-regional model for the simple system described is:
R =
0 F2 F3 0 0 0P1 0 0 M1 0 00 P2 0 0 M2 00 0 0 0 F ′2 F ′3
M ′1 0 0 P ′1 0 0
0 M ′2 0 0 P ′2 0
We can analyze this matrix in the same way we would study any other projectionmatrix
This includes the stable age distribution in the two regions, the reproductive valuesin the two regions, and the growth rate of the overall population
Stanford Summer Short Course: Leslie Matrix II 48
![Page 49: Leslie Matrix II Formal Demography Stanford Spring ... › ... › Jones-Leslie2-050208.pdf · Stanford Summer Short Course: Leslie Matrix II 26. 0 10 20 30 40 0.00 0.05 0.10 0.15](https://reader033.fdocuments.us/reader033/viewer/2022052801/5f17e60145d7940a46529d72/html5/thumbnails/49.jpg)
We can calculate the sensitivities of this growth rate to changes not only in thesurvival probabilities and fertility rates, but also in the migration rates
This might be interesting if, for example, this were a model of rural-urban migrationwhere the urban population has a lower fertility and we wanted to know
Stanford Summer Short Course: Leslie Matrix II 49