SlideShare a Scribd company logo
A Proposal of Two-step Autoregressive Model
Prof. Dr. Loc Nguyen PhD, PostDoc
Loc Nguyen’s Academic Network, Vietnam
Email: ng_phloc@yahoo.com
Homepage: www.locnguyen.net
Two-step Autoregressive Model - Loc Nguyen
14/12/2022 1
Abstract
Autoregressive (AR) model and conditional autoregressive (CAR)
model are specific regressive models in which independent
variables and dependent variable imply the same object. They are
powerful statistical tools to predict values based on correlation of
time domain and space domain, which are useful in epidemiology
analysis. In this research, I combine them by the simple way in
which AR and CAR is estimated in two separate steps so as to
cover time domain and space domain in spatial-temporal data
analysis. Moreover, I integrate logistic model into CAR model,
which aims to improve competence of autoregressive models.
2
Two-step Autoregressive Model - Loc Nguyen
14/12/2022
Table of contents
1. Two-step autoregressive model
2. Integrated with logistic function
3. Conclusions
3
Two-step Autoregressive Model - Loc Nguyen
14/12/2022
1. Two-step autoregressive model
Autoregressive (AR) model evaluates new value of a variable based on its old values; in other
words, independent variables called regressors and dependent variable called responsor refer to the
same concerned object (CO). AR model is often used to predict new values in time series, which is
called temporal autoregressive model. Conditional autoregressive (CAR) model also follows
ideology of AR model where regressors and responsor imply the same CO but CAR model added
more values of neighbors of CO. Therefore, CAR model is often used to predict new values in
space domain, which is called spatial autoregressive model. There are some researches like
(Mariella & Tarantino, 2010) to integrate AR model and CAR model but here I combine them
consequently by the simple way in which AR and CAR is estimated in two separate processes. In
other words, the research uses mainly autoregressive model through two steps:
1. Step 1: CAR model known as spatial autoregressive model is estimated to predict responsor
based on its neighbors and its environmental features like humidity, precipitation, etc. As a
convention, responsor is called case like disease case in epidemiology.
2. Step 2: AR model known as temporal autoregressive model is estimated to forecast next case
based on current cases.
14/12/2022 Two-step Autoregressive Model - Loc Nguyen 4
1. Two-step autoregressive model
Let S = {1, 2,…, n} be spatial domain and so each site s can correspond to an area. Let nb(s) be set of neighbor
sites of site s. Let Xs = (xs1, xs2,…, xsk) be the vector of environmental features at site s and so xsj is the jth
environmental feature at site s. Let Ys be the outcome number of cases at site s. If time is concerned, Xs and Ys
are replaced by Xst and Yst, respectively. Suppose the time interval is T = 2, this research firstly uses spatial
autoregressive model to estimate the number of cases Yst = Ys at current timepoint t. Later on, the research uses
temporal autoregressive model to forecast the future number of cases Ys(t+1) at timepoint t+1 based on current
Yst. Therefore, the computational model of the research is called the two-step autoregressive model.
In the first step, spatial autoregressive model is estimated. Given current timepoint, we have Ys = Yst and Xs
= Xst. When building up spatial autoregressive model, we ignore timepoints in dataset. Because the number of
cases is moved around an average number, Poisson distribution is more suitable to build spatial autoregressive
model. However, for simplicity, I use normal distribution for spatial autoregressive model. Let W be the
symmetric weighted adjacency matrix (Mariella & Tarantino, 2010, p. 227):
𝑊 = 𝑤𝑠𝑠∗ where 𝑤𝑠𝑠∗ =
0 if 𝑠∗
= 𝑠
𝜑 𝑠, 𝑠∗
if 𝑠∗
∈ nb 𝑠
0 otherwise
(1.1)
Where φ(s, s*) measures the approximation of site s and site s* with note that nb(s) denotes the set of neighbors
of site s. Moreover, the set nb(s) does not contain s.
14/12/2022 Two-step Autoregressive Model - Loc Nguyen 5
1. Two-step autoregressive model
Because W is symmetric, we have (Mariella & Tarantino, 2010, p. 227):
∀𝑠, 𝑠∗
∈ 𝑆, 𝑤𝑠𝑠∗ = 𝑤𝑠∗𝑠
Note, W is constant. Let WD be the diagonal matrix of normalization (Mariella & Tarantino, 2010, p.
227):
𝑊𝐷 = diag 𝑤1+, 𝑤2+, … , 𝑤𝑛+
𝑤𝑠+ =
𝑠∗∈nb 𝑠
𝑤𝑠𝑠∗
(1.2)
Let A be the matrix of interaction (Mariella & Tarantino, 2010, p. 227):
𝐴 = 𝑎𝑠𝑠∗ = 𝑊𝐷
−1
𝑊
𝑎𝑠𝑠∗ =
𝑤𝑠𝑠∗
𝑤𝑠+
(1.3)
Note, A is constant. Let ΣD be the diagonal matrix of variances (Mariella & Tarantino, 2010, p. 227):
Σ𝐷 = diag 𝜎1
2
, 𝜎2
2
, … , 𝜎𝑛
2
= 𝜎2
𝑊𝐷
−1
𝜎𝑠
2
=
𝜎2
𝑤𝑠+
(1.4)
Note, σ2 is called global variance.
14/12/2022 Two-step Autoregressive Model - Loc Nguyen 6
1. Two-step autoregressive model
The probability density function (PDF) of the number of cases Ys at Xs given its neighbors Xs* is defined according to normal
distribution as follows (Mariella & Tarantino, 2010, p. 227):
𝑓 𝑌𝑠 𝑋𝑠, 𝑋𝑠∗, 𝛼𝑠, 𝜎2
=
1
2𝜋𝜎𝑠
2
exp −
𝑌𝑠 − 𝛼𝑠
𝑇𝑋𝑠 + 𝜌 𝑠∗∈nb 𝑠 𝑎𝑠𝑠∗ 𝑌𝑠∗ − 𝛼𝑠∗
𝑇
𝑋𝑠∗
2
2𝜎𝑠
2 (1.5)
Where αs = (αs1, αs2,…, αsk) is site regressive coefficient vector and σs
2 is site variance. The constant ρ is called effective
neighborhood constant, |ρ| < 1. Each site variance (local variance) σs
2 was defined.
𝜎𝑠
2
=
𝜎2
𝑤𝑠+
Therefore, only parameters αs and σ2 are parameters in the PDF of Ys at Xs given its neighbors. The mean and variance of Ys
at Xs given its neighbors are:
mean 𝑌𝑠 𝑋𝑠, 𝑋𝑠∗, 𝛼𝑠, 𝜎2
= 𝛼𝑠
𝑇
𝑋𝑠 + 𝜌
𝑠∗∈nb 𝑠
𝑎𝑠𝑠∗ 𝑌𝑠∗ − 𝛼𝑠∗
𝑇
𝑋𝑠∗
var 𝑌𝑠 𝑋𝑠, 𝑋𝑠∗, 𝛼𝑠, 𝜎2 = 𝜎𝑠
2
(1.6)
From the mean of Ys at Xs given its neighbors, the spatial autoregressive model of the number of cases is defined as follows:
𝑌𝑠 = 𝛼𝑠
𝑇𝑋𝑠 + 𝜌
𝑠∗∈nb 𝑠
𝑎𝑠𝑠∗ 𝑌𝑠∗ − 𝛼𝑠∗
𝑇
𝑋𝑠∗ (1.7)
14/12/2022 Two-step Autoregressive Model - Loc Nguyen 7
1. Two-step autoregressive model
The essence of the first step in this research is to estimate all regressive coefficients αs when the
effective neighborhood constant ρ is pre-defined. Let xs
(i) be the ith instance of the vector of
environmental feature Xs at site s. Let ys
(i) be the ith instance of Ys at site s. Let Xs be the set of Ns
instances xs
(i) and let Ys be the set of Ns instances y(i). Each site s owns one dataset (Xs, Ys).
𝑿𝑠 =
𝒙𝑠
1
𝑇
𝒙𝑠
2
𝑇
⋮
𝒙𝑠
𝑁𝑠
𝑇
=
1 𝑥𝑠1
1
𝑥𝑠2
1
⋯ 𝑥𝑠𝑛
1
1 𝑥𝑠1
2
𝑥𝑠2
2
⋯ 𝑥𝑠𝑛
2
⋮ ⋮ ⋮ ⋱ ⋮
1 𝑥𝑠1
𝑁𝑠
𝑥𝑠2
𝑁𝑠
⋯ 𝑥𝑠𝑛
𝑁𝑠
𝒙𝑠
𝑖
=
1
𝑥𝑠1
𝑖
𝑥𝑠2
𝑖
⋮
𝑥𝑠𝑛
𝑖
, 𝒙𝑠
𝑗
=
1
𝑥𝑠1
𝑗
𝑥𝑠2
𝑗
⋮
𝑥𝑠𝑛
𝑗
, 𝒚𝑠 =
𝑦𝑠
1
𝑦𝑠
2
⋮
𝑦𝑠
𝑁𝑠
(1.8)
14/12/2022 Two-step Autoregressive Model - Loc Nguyen 8
1. Two-step autoregressive model
Let X and Y be the sets of all site dataset Xs and Ys, respectively. Let α and σ2 be the sets of all αs and all σs
2,
respectively. The joint probability of the global dataset (X, Y) is:
𝑓 𝑿, 𝒀 𝛼, 𝜎2 =
𝑠=1
𝑆
𝑓 𝑌𝑠 𝑋𝑠, 𝑋𝑠∗, 𝛼𝑠, 𝜎𝑠
2
=
𝑠=1
𝑆
2𝜋𝜎𝑠
2 −
𝑁𝑠
2 ∗
𝑠=1
𝑆
exp −
1
2𝜎𝑠
2
𝑖=1
𝑁𝑠
𝑦𝑠
𝑖
− 𝛼𝑠
𝑇𝒙𝑠
𝑖
+ 𝜌
𝑠∗∈nb 𝑠
𝑎𝑠𝑠∗ 𝑦𝑠∗
𝑖
− 𝛼𝑠∗
𝑇
𝒙𝑠∗
𝑖
2
The log-likelihood function is logarithm function of such joint probability:
𝑙 𝛼, 𝜎2
= −
1
2
𝑠=1
𝑆
𝑁𝑠log 2𝜋𝜎𝑠
2
−
1
2
𝑠=1
𝑆
1
𝜎𝑠
2
𝑖=1
𝑁𝑠
𝑦𝑠
𝑖
− 𝛼𝑠
𝑇
𝒙𝑠
𝑖
+ 𝜌
𝑠∗∈nb 𝑠
𝑎𝑠𝑠∗ 𝑦𝑠∗
𝑖
− 𝛼𝑠∗
𝑇
𝒙𝑠∗
𝑖
2
Let αs
* = (αs1
*, αs2
*,…, αsk
*) be the optimal estimate of regressive coefficients and thus, αs
* is a maximizer of
l(α, σ2).
𝛼𝑠
∗ = argmax
𝛼1,𝛼2,…,𝛼𝑠,…,𝛼 𝑆
𝑙 𝛼, 𝜎2
14/12/2022 Two-step Autoregressive Model - Loc Nguyen 9
1. Two-step autoregressive model
By taking first-order partial derivatives of l(α, σ2) with regard to αs, we obtain:
𝜕𝑙 𝛼, 𝜎2
𝜕𝛼𝑠
=
1
𝜎𝑠
2
𝑖=1
𝑁𝑠
𝑦𝑠
𝑖
− 𝛼𝑠
𝑇𝒙𝑠
𝑖
+ 𝜌
𝑠∗∈nb 𝑠
𝑎𝑠𝑠∗ 𝑦𝑠∗
𝑖
− 𝛼𝑠∗
𝑇
𝒙𝑠∗
𝑖
𝒙𝑠
𝑖
𝑇
When the first-order partial derivatives of l(α, σ2) with regard to αs are equal to zero, it gets locally maximal. In other
words, αs
* is solution of the following equation resulted from setting such derivatives to be zero.
𝑖=1
𝑁𝑠
𝑦𝑠
𝑖
− 𝛼𝑠
𝑇𝒙𝑠
𝑖
+ 𝜌
𝑠∗∈nb 𝑠
𝑎𝑠𝑠∗ 𝑦𝑠∗
𝑖
− 𝛼𝑠∗
𝑇
𝒙𝑠∗
𝑖
𝒙𝑠
𝑖
𝑇
= 𝟎𝑇
As a convention, ass* = 0 if ys*
(i) or xs*
(i) does not exist. The equation above is re-written as follows:
𝒚𝑠
𝑇
𝑿𝑠 − 𝛼𝑠
𝑇
𝑿𝑠
𝑇
𝑿𝑠 − 𝜌
𝑠∗∈nb 𝑠
𝑎𝑠𝑠∗ 𝒚𝑠∗
𝑇
𝑿𝑠 − 𝛼𝑠∗
𝑇
𝑿𝑠
𝑇
𝑿𝑠
= 𝒚𝑠
𝑇 − 𝜌
𝑠∗∈nb 𝑠
𝑎𝑠𝑠∗𝒚𝑠∗
𝑇
𝑿𝑠 − 𝛼𝑠
𝑇 − 𝜌
𝑠∗∈nb 𝑠
𝑎𝑠𝑠∗𝛼𝑠∗
𝑇
𝑿𝑠
𝑇𝑿𝑠 = 𝟎𝑇
This implies
𝛼𝑠 − 𝜌
𝑠∗∈nb 𝑠
𝑎𝑠𝑠∗𝛼𝑠∗ = 𝑿𝑠
𝑇𝑿𝑠
−1𝑿𝑠
𝑇 𝒚𝑠 − 𝜌
𝑠∗∈nb 𝑠
𝑎𝑠𝑠∗𝒚𝑠∗
14/12/2022 Two-step Autoregressive Model - Loc Nguyen 10
1. Two-step autoregressive model
In general, the estimates αs
* for all regressive coefficient vectors αs where s from
1 to |S| is solution of the linear equation system consisting of |S| equations as
follows:
𝛼𝑠 − 𝜌
𝑠∗∈𝑛𝑏 𝑠
𝑎𝑠𝑠∗𝛼𝑠∗ = 𝑿𝑠
𝑇
𝑿𝑠
−1
𝑿𝑠
𝑇
𝒚𝑠 − 𝜌
𝑠∗∈𝑛𝑏 𝑠
𝑎𝑠𝑠∗𝒚𝑠∗
𝑠=1, 𝑆
(1.9)
It is not difficult to solve the linear equation system above. As a result, the
spatial autoregressive model is re-written as follows:
𝑌𝑠 = 𝛼𝑠
∗ 𝑇
𝑋𝑠 + 𝜌
𝑠∗∈nb 𝑠
𝑎𝑠𝑠∗ 𝑌𝑠∗ − 𝛼𝑠∗
∗ 𝑇
𝑋𝑠∗ (1.10)
14/12/2022 Two-step Autoregressive Model - Loc Nguyen 11
1. Two-step autoregressive model
In the second step, temporal autoregressive model is estimated. For simplicity, we follow the Markov
condition in which, future cases only depend on current cases. When building up spatial autoregressive
model, we ignore environmental features Xs in dataset or we focus on a given site s. Suppose the current
number of cases Yst at site s was determined previously by spatial autoregressive model. As a convention,
the “s” index is removed from the response variable and so we have Yt+1 = Ys(t+1) and Yt = Yst. We will
forecast next (unknown) Yt+1 based on current (known) Yt. The temporal autoregressive model is defined as
follows (Eshel, 2003, p. 1):
𝑌𝑡+1 = 𝛽𝑌𝑡 + 𝜀 (1.11)
Where ε is the error random variable and β is temporal coefficient with note that mean of ε is 0 such that
𝐸 𝑌𝑡+1 = 𝛽𝑌𝑡
Let yt be the value of Yt at timepoint t. Suppose there are T timepoints and we have t = 1, 2,…, T. The sum
of squared errors is defined as follows:
𝑆 𝛽 =
𝑡=1
𝑇−1
𝑦𝑡+1 − 𝛽𝑦𝑡
2
Let β* be the estimate of β. The least squares method sets β* as the minimizer of S(β).
𝛽∗
= argmin
𝛽
𝑆 𝛽
14/12/2022 Two-step Autoregressive Model - Loc Nguyen 12
1. Two-step autoregressive model
By taking first-order derivative of S(β) with regard to β, we obtain (Eshel, 2003, p. 1):
𝑑𝑆 𝛽
𝑑𝛽
= −2
𝑡=1
𝑇−1
𝑦𝑡+1 − 𝛽𝑦𝑡 𝑦𝑡
The temporal coefficient β* is solution of the following equation resulted from setting such
derivative to be zero.
−2
𝑡=1
𝑇−1
𝑦𝑡+1 − 𝛽𝑦𝑡 𝑦𝑡 = 0
It is easy to determine the formula for calculating β* (Eshel, 2003, p. 1).
𝛽∗ =
𝑡=1
𝑇−1
𝑦𝑡𝑦𝑡+1
𝑡=1
𝑇−1
𝑦𝑡
2 (1.12)
Please pay attention that values yt are calculated by CAR model estimated in the first step. The
temporal autoregressive model is re-written as follows:
𝑌𝑠 𝑡+1 = 𝛽∗𝑌𝑠𝑡 (1.13)
Equation 1.12 to calculate β* is simplest form of the well-known Yule-Walker equation.
14/12/2022 Two-step Autoregressive Model - Loc Nguyen 13
2. Integrated with logistic function
In normal CAR model, the responsor Ys which is the outcome number of cases at site s is real variable but in the situation
that Ys is category variable whose G ordered values are v1, v2,…, vG, new modification of CAR model is required. The
multinomial logistic function of Ys with regard to value vg at site s, which is associated spatial autoregression, is proposed
as follows:
𝑓𝑔 𝑌𝑠 𝑋𝑠, 𝑋𝑠∗, 𝛼𝑠𝑔 =
1
1 + exp − 𝛼𝑠𝑔
𝑇
𝑋𝑠 + 𝜌 𝑠∗∈nb 𝑠 𝑎𝑠𝑠∗ 𝛼𝑠𝑔
𝑇
𝑋𝑠∗ − 𝛼𝑠∗𝑔
𝑇
𝑋𝑠∗
(2.1)
Note that αsg = (αsg1, αsg2,…, αsgk) which is site regressive coefficient vector with regard to value yg is the main parameter
which is estimated here. Let pg be the probability that Ys receives value yg, where g is from 1 to G. If G values v1, v2,…, vG
are not ordered, pg is the same to the proposed multinomial logistic function of Ys.
𝑝 𝑋𝑠 𝛼𝑠𝑔 = 𝑃 𝑌𝑠 = 𝑣𝑔 𝑋𝑠, 𝑋𝑠∗, 𝛼𝑠𝑔 = 𝑓𝑔 𝑌𝑠 𝑋𝑠, 𝑋𝑠∗, 𝛼𝑠𝑔
=
1
1 + exp − 𝛼𝑠𝑔
𝑇
𝑋𝑠 + 𝜌 𝑠∗∈nb 𝑠 𝑎𝑠𝑠∗ 𝛼𝑠𝑔
𝑇
𝑋𝑠∗ − 𝛼𝑠∗𝑔
𝑇
𝑋𝑠∗
(2.2)
If G values v1, v2,…, vG are ordered such that v1 ≤ v2 ≤ … ≤ vG then, according to Bender and Grouven (Bender & Grouven,
1997, p. 547), pg becomes the cumulative probability such that Ys ≤ vg as follows:
𝑝 𝑋𝑠 𝛼𝑠𝑔 = 𝑃 𝑌𝑠 ≤ 𝑣𝑔 𝑋𝑠, 𝑋𝑠∗, 𝛼𝑠𝑔 =
𝑖=1
𝑔
𝑓𝑖 𝑌𝑠 𝑋𝑠, 𝑋𝑠∗, 𝛼𝑠𝑖 (2.3)
Obviously, p(Xs|αsg) is the CAR model integrated with logistic function, which is called LCAR model.
14/12/2022 Two-step Autoregressive Model - Loc Nguyen 14
2. Integrated with logistic function
Essentially, LCAR is determined by estimating parameters αsg. As a convention, let
𝑝𝑠𝑔 = 𝑝 𝑋𝑠 𝛼𝑠𝑔
𝑔=1
𝐺
𝑝𝑠𝑔 = 1
(2.4)
The logarithm of psg which is called logit of psg is defined as follows:
logit 𝑝𝑠𝑔 = log
𝑝𝑠𝑔
1 − 𝑝𝑠𝑔
= 𝛼𝑠𝑔
𝑇
𝑋𝑠 + 𝜌
𝑠∗∈nb 𝑠
𝑎𝑠𝑠∗ 𝛼𝑠𝑔
𝑇
𝑋𝑠∗ − 𝛼𝑠∗𝑔
𝑇
𝑋𝑠∗ (2.5)
Suppose Ys obeys multinomial distribution so that the parameters αsg at site s are determined by maximum likelihood
estimation (MLE) method (Czepiel, 2002). The joint probability of the global dataset (X, Y) in reduced form is:
𝑓 𝑿, 𝒀 𝛼 =
𝑠=1
𝑆
𝑔=1
𝐺
𝑖=1,𝑦𝑠
𝑖
=𝑣𝑔
𝑁𝑠
𝑝𝑠𝑔
𝑖
Where psg
(i) is the evaluation of psg with the instance Xs = xs
(i). The log-likelihood function is logarithm function of such
joint probability:
𝑙 𝛼 = −
𝑠=1
𝑆
𝑔=1
𝐺
𝑖=1,𝑦𝑠
𝑖
=𝑣𝑔
𝑁𝑠
log 1 + exp − 𝛼𝑠𝑔
𝑇
𝒙𝑠
𝑖
+ 𝜌
𝑠∗∈nb 𝑠
𝑎𝑠𝑠∗ 𝛼𝑠𝑔
𝑇
𝒙𝑠∗
𝑖
− 𝛼𝑠∗𝑔
𝑇
𝒙𝑠∗
𝑖
14/12/2022 Two-step Autoregressive Model - Loc Nguyen 15
2. Integrated with logistic function
Let αsg
* = (αsg1
*, αsg2
*,…, αsgk
*) be the optimal estimate of regressive coefficients which is a maximizer of l(α)
with the constraint 𝑔=1
𝐺
𝑝𝑠𝑔 = 1 which is equivalent to:
log
𝑔=1
𝐺
𝑝𝑠𝑔 = 0
Let L0(α, λ) be Lagrange function defined as follows:
𝐿0 𝛼, 𝜆 = 𝑙 𝛼 +
𝑠=1
𝑆
𝑔=1
𝐺
𝑖=1,𝑦𝑠
𝑖
=𝑣𝑔
𝑁𝑠
𝜆𝑖
𝑠𝑔
log 𝑝𝑠𝑔
𝑖
Where λi are Lagrange multipliers such that λi ≥ 0 with note that each λi
(sg) implies λi at site s given value vg.
Because logarithm is concave function, we apply Jessen inequation to approximate the Lagrange function as
follows:
𝐿0 𝛼, 𝜆 ≤ 𝐿 𝛼, 𝜆 = 𝑙 𝛼 +
𝑠=1
𝑆
𝑔=1
𝐺
𝑖=1,𝑦𝑠
𝑖
=𝑣𝑔
𝑁𝑠
𝜆𝑖
𝑠𝑔
log 𝑝𝑠𝑔
𝑖
We now focus on maximizing L(α, λ).
14/12/2022 Two-step Autoregressive Model - Loc Nguyen 16
2. Integrated with logistic function
We now focus on maximizing L(α, λ). The first-order partial derivative of L(α, λ) with regard to αsg is:
𝜕𝐿 𝛼, 𝜆
𝜕𝛼𝑠𝑔
=
𝑖=1,𝑦𝑠
𝑖
=𝑣𝑔
𝑁𝑠
𝜆𝑖 + 1 1 − 𝑝𝑠𝑔
𝑖
𝒙𝑠
𝑖
𝑇
+ 𝜌
𝑠∗∈nb 𝑠
𝑎𝑠𝑠∗ 𝒙𝑠∗
𝑖
𝑇
Note, ass* = 0 if ys*
(i) or xs*
(i) does not exist. Let
𝑍𝑠
𝑖
= 𝒙𝑠
𝑖
+ 𝜌
𝑠∗∈nb 𝑠
𝑎𝑠𝑠∗𝒙𝑠∗
𝑖
(2.6)
We have:
𝜕𝐿 𝛼, 𝜆
𝜕𝛼𝑠𝑔
=
𝑖=1,𝑦𝑠
𝑖
=𝑣𝑔
𝑁𝑠
𝜆𝑖 + 1 1 − 𝑝𝑠𝑔
𝑖
𝑍𝑠
𝑖
𝑇
Setting the first-order partial derivative of L(α, λ) with regard to λi to be 0, we obtain:
𝜕𝐿 𝛼, 𝜆
𝜕𝜆𝑖
= 𝜆𝑖
𝑔=1
𝐺
log 𝑝𝑠𝑔
𝑖
= 0 ⇒ 𝜆𝑖 = 0
Substituting all λi = 0 into the derivative
𝜕𝐿 𝛼,𝜆
𝜕𝛼𝑠𝑔
, we have:
𝜕𝐿 𝛼, 𝜆
𝜕𝛼𝑠𝑔
=
𝑖=1,𝑦𝑠
𝑖
=𝑣𝑔
𝑁𝑠
1 − 𝑝𝑠𝑔
𝑖
𝑍𝑠
𝑖
𝑇
14/12/2022 Two-step Autoregressive Model - Loc Nguyen 17
2. Integrated with logistic function
By setting such derivative
𝜕𝐿 𝛼,𝜆
𝜕𝛼𝑠𝑔
to be zero, we obtain:
𝑖=1,𝑦𝑠
𝑖
=𝑣𝑔
𝑁𝑠
𝑝𝑠𝑔
𝑖
𝑍𝑠
𝑖
=
𝑖=1,𝑦𝑠
𝑖
=𝑣𝑔
𝑁𝑠
𝑍𝑠
𝑖
As a result, the estimates αsg
* for all regressive coefficient vectors αsg over all sites and all
values v1, v2,…, vG is solution of the linear equation system consisting of |S|G equations as
follows:
𝑖=1,𝑦𝑠
𝑖
=𝑣𝑔
𝑁𝑠
𝑝𝑠𝑔
𝑖
𝑍𝑠
𝑖
=
𝑖=1,𝑦𝑠
𝑖
=𝑣𝑔
𝑁𝑠
𝑍𝑠
𝑖
𝑠=1, 𝑆 ,𝑔=1,𝐺
(2.7)
In general, the equation 2.7 is used to estimate LCAR model, which can be solved by
numerical methods like Newton-Raphson method for nonlinear equation system (Burden &
Faires, 2011, pp. 638-642).
14/12/2022 Two-step Autoregressive Model - Loc Nguyen 18
3. Conclusions
In this research, I proposed a simple integration model for spatial-
temporal autoregression which consists of two consequent and
separated steps. Moreover, I also proposed the integration of
logistic model and autoregressive model called LCAR which is
slightly different from traditional CAR model. However, the
traditional AR model known as temporal autoregressive model is
still simple with assumption of first-order Markov property. In the
future trend, I will improve AR model with autoregressive moving
average (ARMA) and autoregressive integrated moving average
(ARIMA).
14/12/2022 Two-step Autoregressive Model - Loc Nguyen 19
References
1. Bender, R., & Grouven, U. (1997, September 5). Ordinal logistic regression in medical
research. Journal of the Royal College of Physicians of London, 31(5), 546-551.
Retrieved from https://pubmed.ncbi.nlm.nih.gov/9429194/
2. Burden, R. L., & Faires, D. J. (2011). Numerical Analysis (9th Edition ed.). (M. Julet,
Ed.) Brooks/Cole Cengage Learning.
3. Czepiel, S. A. (2002). Maximum Likelihood Estimation of Logistic Regression Models:
Theory and Implementation. Czepiel's website http://czep.net.
4. Eshel, G. (2003). The Yule Walker Equations for the AR Coefficients. University of
Wharton, Statistics Department. Michael Steele Homepage. Retrieved 11 24, 2018,
from http://www-
stat.wharton.upenn.edu/~steele/Courses/956/ResourceDetails/YWSourceFiles/YW-
Eshel.pdf
5. Mariella, L., & Tarantino, M. (2010). Spatial Temporal Conditional Auto-Regressive
Model: A New Autoregressive Matrix. Austrian Journal of Statistics, 39(3), 223-244.
doi:10.17713/ajs.v39i3.246
14/12/2022 Two-step Autoregressive Model - Loc Nguyen 20
Thank you for listening
21
Two-step Autoregressive Model - Loc Nguyen
14/12/2022

More Related Content

Similar to A Proposal of Two-step Autoregressive Model

QTML2021 UAP Quantum Feature Map
QTML2021 UAP Quantum Feature MapQTML2021 UAP Quantum Feature Map
QTML2021 UAP Quantum Feature Map
Ha Phuong
 
Imecs2012 pp440 445
Imecs2012 pp440 445Imecs2012 pp440 445
Imecs2012 pp440 445
Rasha Orban
 
Ijetr021210
Ijetr021210Ijetr021210
Ijetr021210
Ijetr021210Ijetr021210
Ijetr021210
ER Publication.org
 
Normalized averaging using adaptive applicability functions with applications...
Normalized averaging using adaptive applicability functions with applications...Normalized averaging using adaptive applicability functions with applications...
Normalized averaging using adaptive applicability functions with applications...
guest31063e
 
Erwin. e. obermayer k._schulten. k. _1992: self-organising maps_stationary st...
Erwin. e. obermayer k._schulten. k. _1992: self-organising maps_stationary st...Erwin. e. obermayer k._schulten. k. _1992: self-organising maps_stationary st...
Erwin. e. obermayer k._schulten. k. _1992: self-organising maps_stationary st...
ArchiLab 7
 
International Journal of Engineering Research and Development
International Journal of Engineering Research and DevelopmentInternational Journal of Engineering Research and Development
International Journal of Engineering Research and Development
IJERD Editor
 
Parellelism in spectral methods
Parellelism in spectral methodsParellelism in spectral methods
Parellelism in spectral methods
Ramona Corman
 
Robust Fuzzy Data Clustering In An Ordinal Scale Based On A Similarity Measure
Robust Fuzzy Data Clustering In An Ordinal Scale Based On A Similarity MeasureRobust Fuzzy Data Clustering In An Ordinal Scale Based On A Similarity Measure
Robust Fuzzy Data Clustering In An Ordinal Scale Based On A Similarity Measure
IJRES Journal
 
Pakdd
PakddPakdd
Pakdd
Siswanto .
 
first research paper
first research paperfirst research paper
first research paper
Justin McKennon
 
Dynamics
DynamicsDynamics
Dynamics
nguyendattdh
 
Functional analysis in mechanics
Functional analysis in mechanicsFunctional analysis in mechanics
Functional analysis in mechanics
Springer
 
Functional analysis in mechanics 2e
Functional analysis in mechanics  2eFunctional analysis in mechanics  2e
Functional analysis in mechanics 2e
Springer
 
BNL_Research_Report
BNL_Research_ReportBNL_Research_Report
BNL_Research_Report
Brandon McKinzie
 
A COMPARISON OF PARTICLE SWARM OPTIMIZATION AND DIFFERENTIAL EVOLUTION
A COMPARISON OF PARTICLE SWARM OPTIMIZATION AND DIFFERENTIAL EVOLUTIONA COMPARISON OF PARTICLE SWARM OPTIMIZATION AND DIFFERENTIAL EVOLUTION
A COMPARISON OF PARTICLE SWARM OPTIMIZATION AND DIFFERENTIAL EVOLUTION
ijsc
 
lcr
lcrlcr
Wang1998
Wang1998Wang1998
Adaptive filtersfinal
Adaptive filtersfinalAdaptive filtersfinal
Adaptive filtersfinal
Wiw Miu
 
Dycops2019
Dycops2019 Dycops2019
Dycops2019
Jéssyca Bessa
 

Similar to A Proposal of Two-step Autoregressive Model (20)

QTML2021 UAP Quantum Feature Map
QTML2021 UAP Quantum Feature MapQTML2021 UAP Quantum Feature Map
QTML2021 UAP Quantum Feature Map
 
Imecs2012 pp440 445
Imecs2012 pp440 445Imecs2012 pp440 445
Imecs2012 pp440 445
 
Ijetr021210
Ijetr021210Ijetr021210
Ijetr021210
 
Ijetr021210
Ijetr021210Ijetr021210
Ijetr021210
 
Normalized averaging using adaptive applicability functions with applications...
Normalized averaging using adaptive applicability functions with applications...Normalized averaging using adaptive applicability functions with applications...
Normalized averaging using adaptive applicability functions with applications...
 
Erwin. e. obermayer k._schulten. k. _1992: self-organising maps_stationary st...
Erwin. e. obermayer k._schulten. k. _1992: self-organising maps_stationary st...Erwin. e. obermayer k._schulten. k. _1992: self-organising maps_stationary st...
Erwin. e. obermayer k._schulten. k. _1992: self-organising maps_stationary st...
 
International Journal of Engineering Research and Development
International Journal of Engineering Research and DevelopmentInternational Journal of Engineering Research and Development
International Journal of Engineering Research and Development
 
Parellelism in spectral methods
Parellelism in spectral methodsParellelism in spectral methods
Parellelism in spectral methods
 
Robust Fuzzy Data Clustering In An Ordinal Scale Based On A Similarity Measure
Robust Fuzzy Data Clustering In An Ordinal Scale Based On A Similarity MeasureRobust Fuzzy Data Clustering In An Ordinal Scale Based On A Similarity Measure
Robust Fuzzy Data Clustering In An Ordinal Scale Based On A Similarity Measure
 
Pakdd
PakddPakdd
Pakdd
 
first research paper
first research paperfirst research paper
first research paper
 
Dynamics
DynamicsDynamics
Dynamics
 
Functional analysis in mechanics
Functional analysis in mechanicsFunctional analysis in mechanics
Functional analysis in mechanics
 
Functional analysis in mechanics 2e
Functional analysis in mechanics  2eFunctional analysis in mechanics  2e
Functional analysis in mechanics 2e
 
BNL_Research_Report
BNL_Research_ReportBNL_Research_Report
BNL_Research_Report
 
A COMPARISON OF PARTICLE SWARM OPTIMIZATION AND DIFFERENTIAL EVOLUTION
A COMPARISON OF PARTICLE SWARM OPTIMIZATION AND DIFFERENTIAL EVOLUTIONA COMPARISON OF PARTICLE SWARM OPTIMIZATION AND DIFFERENTIAL EVOLUTION
A COMPARISON OF PARTICLE SWARM OPTIMIZATION AND DIFFERENTIAL EVOLUTION
 
lcr
lcrlcr
lcr
 
Wang1998
Wang1998Wang1998
Wang1998
 
Adaptive filtersfinal
Adaptive filtersfinalAdaptive filtersfinal
Adaptive filtersfinal
 
Dycops2019
Dycops2019 Dycops2019
Dycops2019
 

More from Loc Nguyen

Adversarial Variational Autoencoders to extend and improve generative model -...
Adversarial Variational Autoencoders to extend and improve generative model -...Adversarial Variational Autoencoders to extend and improve generative model -...
Adversarial Variational Autoencoders to extend and improve generative model -...
Loc Nguyen
 
Conditional mixture model and its application for regression model
Conditional mixture model and its application for regression modelConditional mixture model and its application for regression model
Conditional mixture model and its application for regression model
Loc Nguyen
 
Nghịch dân chủ luận (tổng quan về dân chủ và thể chế chính trị liên quan đến ...
Nghịch dân chủ luận (tổng quan về dân chủ và thể chế chính trị liên quan đến ...Nghịch dân chủ luận (tổng quan về dân chủ và thể chế chính trị liên quan đến ...
Nghịch dân chủ luận (tổng quan về dân chủ và thể chế chính trị liên quan đến ...
Loc Nguyen
 
A Novel Collaborative Filtering Algorithm by Bit Mining Frequent Itemsets
A Novel Collaborative Filtering Algorithm by Bit Mining Frequent ItemsetsA Novel Collaborative Filtering Algorithm by Bit Mining Frequent Itemsets
A Novel Collaborative Filtering Algorithm by Bit Mining Frequent Itemsets
Loc Nguyen
 
Simple image deconvolution based on reverse image convolution and backpropaga...
Simple image deconvolution based on reverse image convolution and backpropaga...Simple image deconvolution based on reverse image convolution and backpropaga...
Simple image deconvolution based on reverse image convolution and backpropaga...
Loc Nguyen
 
Technological Accessibility: Learning Platform Among Senior High School Students
Technological Accessibility: Learning Platform Among Senior High School StudentsTechnological Accessibility: Learning Platform Among Senior High School Students
Technological Accessibility: Learning Platform Among Senior High School Students
Loc Nguyen
 
Engineering for Social Impact
Engineering for Social ImpactEngineering for Social Impact
Engineering for Social Impact
Loc Nguyen
 
Harnessing Technology for Research Education
Harnessing Technology for Research EducationHarnessing Technology for Research Education
Harnessing Technology for Research Education
Loc Nguyen
 
Future of education with support of technology
Future of education with support of technologyFuture of education with support of technology
Future of education with support of technology
Loc Nguyen
 
Where the dragon to fly
Where the dragon to flyWhere the dragon to fly
Where the dragon to fly
Loc Nguyen
 
Adversarial Variational Autoencoders to extend and improve generative model
Adversarial Variational Autoencoders to extend and improve generative modelAdversarial Variational Autoencoders to extend and improve generative model
Adversarial Variational Autoencoders to extend and improve generative model
Loc Nguyen
 
Learning dyadic data and predicting unaccomplished co-occurrent values by mix...
Learning dyadic data and predicting unaccomplished co-occurrent values by mix...Learning dyadic data and predicting unaccomplished co-occurrent values by mix...
Learning dyadic data and predicting unaccomplished co-occurrent values by mix...
Loc Nguyen
 
Tutorial on Bayesian optimization
Tutorial on Bayesian optimizationTutorial on Bayesian optimization
Tutorial on Bayesian optimization
Loc Nguyen
 
Tutorial on Support Vector Machine
Tutorial on Support Vector MachineTutorial on Support Vector Machine
Tutorial on Support Vector Machine
Loc Nguyen
 
Jagged stock investment strategy
Jagged stock investment strategyJagged stock investment strategy
Jagged stock investment strategy
Loc Nguyen
 
A short study on minima distribution
A short study on minima distributionA short study on minima distribution
A short study on minima distribution
Loc Nguyen
 
Tutorial on EM algorithm – Part 4
Tutorial on EM algorithm – Part 4Tutorial on EM algorithm – Part 4
Tutorial on EM algorithm – Part 4
Loc Nguyen
 
Tutorial on EM algorithm – Part 3
Tutorial on EM algorithm – Part 3Tutorial on EM algorithm – Part 3
Tutorial on EM algorithm – Part 3
Loc Nguyen
 
Tutorial on particle swarm optimization
Tutorial on particle swarm optimizationTutorial on particle swarm optimization
Tutorial on particle swarm optimization
Loc Nguyen
 
Tutorial on EM algorithm - Poster
Tutorial on EM algorithm - PosterTutorial on EM algorithm - Poster
Tutorial on EM algorithm - Poster
Loc Nguyen
 

More from Loc Nguyen (20)

Adversarial Variational Autoencoders to extend and improve generative model -...
Adversarial Variational Autoencoders to extend and improve generative model -...Adversarial Variational Autoencoders to extend and improve generative model -...
Adversarial Variational Autoencoders to extend and improve generative model -...
 
Conditional mixture model and its application for regression model
Conditional mixture model and its application for regression modelConditional mixture model and its application for regression model
Conditional mixture model and its application for regression model
 
Nghịch dân chủ luận (tổng quan về dân chủ và thể chế chính trị liên quan đến ...
Nghịch dân chủ luận (tổng quan về dân chủ và thể chế chính trị liên quan đến ...Nghịch dân chủ luận (tổng quan về dân chủ và thể chế chính trị liên quan đến ...
Nghịch dân chủ luận (tổng quan về dân chủ và thể chế chính trị liên quan đến ...
 
A Novel Collaborative Filtering Algorithm by Bit Mining Frequent Itemsets
A Novel Collaborative Filtering Algorithm by Bit Mining Frequent ItemsetsA Novel Collaborative Filtering Algorithm by Bit Mining Frequent Itemsets
A Novel Collaborative Filtering Algorithm by Bit Mining Frequent Itemsets
 
Simple image deconvolution based on reverse image convolution and backpropaga...
Simple image deconvolution based on reverse image convolution and backpropaga...Simple image deconvolution based on reverse image convolution and backpropaga...
Simple image deconvolution based on reverse image convolution and backpropaga...
 
Technological Accessibility: Learning Platform Among Senior High School Students
Technological Accessibility: Learning Platform Among Senior High School StudentsTechnological Accessibility: Learning Platform Among Senior High School Students
Technological Accessibility: Learning Platform Among Senior High School Students
 
Engineering for Social Impact
Engineering for Social ImpactEngineering for Social Impact
Engineering for Social Impact
 
Harnessing Technology for Research Education
Harnessing Technology for Research EducationHarnessing Technology for Research Education
Harnessing Technology for Research Education
 
Future of education with support of technology
Future of education with support of technologyFuture of education with support of technology
Future of education with support of technology
 
Where the dragon to fly
Where the dragon to flyWhere the dragon to fly
Where the dragon to fly
 
Adversarial Variational Autoencoders to extend and improve generative model
Adversarial Variational Autoencoders to extend and improve generative modelAdversarial Variational Autoencoders to extend and improve generative model
Adversarial Variational Autoencoders to extend and improve generative model
 
Learning dyadic data and predicting unaccomplished co-occurrent values by mix...
Learning dyadic data and predicting unaccomplished co-occurrent values by mix...Learning dyadic data and predicting unaccomplished co-occurrent values by mix...
Learning dyadic data and predicting unaccomplished co-occurrent values by mix...
 
Tutorial on Bayesian optimization
Tutorial on Bayesian optimizationTutorial on Bayesian optimization
Tutorial on Bayesian optimization
 
Tutorial on Support Vector Machine
Tutorial on Support Vector MachineTutorial on Support Vector Machine
Tutorial on Support Vector Machine
 
Jagged stock investment strategy
Jagged stock investment strategyJagged stock investment strategy
Jagged stock investment strategy
 
A short study on minima distribution
A short study on minima distributionA short study on minima distribution
A short study on minima distribution
 
Tutorial on EM algorithm – Part 4
Tutorial on EM algorithm – Part 4Tutorial on EM algorithm – Part 4
Tutorial on EM algorithm – Part 4
 
Tutorial on EM algorithm – Part 3
Tutorial on EM algorithm – Part 3Tutorial on EM algorithm – Part 3
Tutorial on EM algorithm – Part 3
 
Tutorial on particle swarm optimization
Tutorial on particle swarm optimizationTutorial on particle swarm optimization
Tutorial on particle swarm optimization
 
Tutorial on EM algorithm - Poster
Tutorial on EM algorithm - PosterTutorial on EM algorithm - Poster
Tutorial on EM algorithm - Poster
 

Recently uploaded

aziz sancar nobel prize winner: from mardin to nobel
aziz sancar nobel prize winner: from mardin to nobelaziz sancar nobel prize winner: from mardin to nobel
aziz sancar nobel prize winner: from mardin to nobel
İsa Badur
 
ANAMOLOUS SECONDARY GROWTH IN DICOT ROOTS.pptx
ANAMOLOUS SECONDARY GROWTH IN DICOT ROOTS.pptxANAMOLOUS SECONDARY GROWTH IN DICOT ROOTS.pptx
ANAMOLOUS SECONDARY GROWTH IN DICOT ROOTS.pptx
RASHMI M G
 
Deep Software Variability and Frictionless Reproducibility
Deep Software Variability and Frictionless ReproducibilityDeep Software Variability and Frictionless Reproducibility
Deep Software Variability and Frictionless Reproducibility
University of Rennes, INSA Rennes, Inria/IRISA, CNRS
 
Randomised Optimisation Algorithms in DAPHNE
Randomised Optimisation Algorithms in DAPHNERandomised Optimisation Algorithms in DAPHNE
Randomised Optimisation Algorithms in DAPHNE
University of Maribor
 
3D Hybrid PIC simulation of the plasma expansion (ISSS-14)
3D Hybrid PIC simulation of the plasma expansion (ISSS-14)3D Hybrid PIC simulation of the plasma expansion (ISSS-14)
3D Hybrid PIC simulation of the plasma expansion (ISSS-14)
David Osipyan
 
mô tả các thí nghiệm về đánh giá tác động dòng khí hóa sau đốt
mô tả các thí nghiệm về đánh giá tác động dòng khí hóa sau đốtmô tả các thí nghiệm về đánh giá tác động dòng khí hóa sau đốt
mô tả các thí nghiệm về đánh giá tác động dòng khí hóa sau đốt
HongcNguyn6
 
ESR spectroscopy in liquid food and beverages.pptx
ESR spectroscopy in liquid food and beverages.pptxESR spectroscopy in liquid food and beverages.pptx
ESR spectroscopy in liquid food and beverages.pptx
PRIYANKA PATEL
 
EWOCS-I: The catalog of X-ray sources in Westerlund 1 from the Extended Weste...
EWOCS-I: The catalog of X-ray sources in Westerlund 1 from the Extended Weste...EWOCS-I: The catalog of X-ray sources in Westerlund 1 from the Extended Weste...
EWOCS-I: The catalog of X-ray sources in Westerlund 1 from the Extended Weste...
Sérgio Sacani
 
Comparing Evolved Extractive Text Summary Scores of Bidirectional Encoder Rep...
Comparing Evolved Extractive Text Summary Scores of Bidirectional Encoder Rep...Comparing Evolved Extractive Text Summary Scores of Bidirectional Encoder Rep...
Comparing Evolved Extractive Text Summary Scores of Bidirectional Encoder Rep...
University of Maribor
 
Shallowest Oil Discovery of Turkiye.pptx
Shallowest Oil Discovery of Turkiye.pptxShallowest Oil Discovery of Turkiye.pptx
Shallowest Oil Discovery of Turkiye.pptx
Gokturk Mehmet Dilci
 
Bob Reedy - Nitrate in Texas Groundwater.pdf
Bob Reedy - Nitrate in Texas Groundwater.pdfBob Reedy - Nitrate in Texas Groundwater.pdf
Bob Reedy - Nitrate in Texas Groundwater.pdf
Texas Alliance of Groundwater Districts
 
Topic: SICKLE CELL DISEASE IN CHILDREN-3.pdf
Topic: SICKLE CELL DISEASE IN CHILDREN-3.pdfTopic: SICKLE CELL DISEASE IN CHILDREN-3.pdf
Topic: SICKLE CELL DISEASE IN CHILDREN-3.pdf
TinyAnderson
 
Nucleophilic Addition of carbonyl compounds.pptx
Nucleophilic Addition of carbonyl  compounds.pptxNucleophilic Addition of carbonyl  compounds.pptx
Nucleophilic Addition of carbonyl compounds.pptx
SSR02
 
原版制作(carleton毕业证书)卡尔顿大学毕业证硕士文凭原版一模一样
原版制作(carleton毕业证书)卡尔顿大学毕业证硕士文凭原版一模一样原版制作(carleton毕业证书)卡尔顿大学毕业证硕士文凭原版一模一样
原版制作(carleton毕业证书)卡尔顿大学毕业证硕士文凭原版一模一样
yqqaatn0
 
Applied Science: Thermodynamics, Laws & Methodology.pdf
Applied Science: Thermodynamics, Laws & Methodology.pdfApplied Science: Thermodynamics, Laws & Methodology.pdf
Applied Science: Thermodynamics, Laws & Methodology.pdf
University of Hertfordshire
 
bordetella pertussis.................................ppt
bordetella pertussis.................................pptbordetella pertussis.................................ppt
bordetella pertussis.................................ppt
kejapriya1
 
Chapter 12 - climate change and the energy crisis
Chapter 12 - climate change and the energy crisisChapter 12 - climate change and the energy crisis
Chapter 12 - climate change and the energy crisis
tonzsalvador2222
 
SAR of Medicinal Chemistry 1st by dk.pdf
SAR of Medicinal Chemistry 1st by dk.pdfSAR of Medicinal Chemistry 1st by dk.pdf
SAR of Medicinal Chemistry 1st by dk.pdf
KrushnaDarade1
 
Phenomics assisted breeding in crop improvement
Phenomics assisted breeding in crop improvementPhenomics assisted breeding in crop improvement
Phenomics assisted breeding in crop improvement
IshaGoswami9
 
Nucleic Acid-its structural and functional complexity.
Nucleic Acid-its structural and functional complexity.Nucleic Acid-its structural and functional complexity.
Nucleic Acid-its structural and functional complexity.
Nistarini College, Purulia (W.B) India
 

Recently uploaded (20)

aziz sancar nobel prize winner: from mardin to nobel
aziz sancar nobel prize winner: from mardin to nobelaziz sancar nobel prize winner: from mardin to nobel
aziz sancar nobel prize winner: from mardin to nobel
 
ANAMOLOUS SECONDARY GROWTH IN DICOT ROOTS.pptx
ANAMOLOUS SECONDARY GROWTH IN DICOT ROOTS.pptxANAMOLOUS SECONDARY GROWTH IN DICOT ROOTS.pptx
ANAMOLOUS SECONDARY GROWTH IN DICOT ROOTS.pptx
 
Deep Software Variability and Frictionless Reproducibility
Deep Software Variability and Frictionless ReproducibilityDeep Software Variability and Frictionless Reproducibility
Deep Software Variability and Frictionless Reproducibility
 
Randomised Optimisation Algorithms in DAPHNE
Randomised Optimisation Algorithms in DAPHNERandomised Optimisation Algorithms in DAPHNE
Randomised Optimisation Algorithms in DAPHNE
 
3D Hybrid PIC simulation of the plasma expansion (ISSS-14)
3D Hybrid PIC simulation of the plasma expansion (ISSS-14)3D Hybrid PIC simulation of the plasma expansion (ISSS-14)
3D Hybrid PIC simulation of the plasma expansion (ISSS-14)
 
mô tả các thí nghiệm về đánh giá tác động dòng khí hóa sau đốt
mô tả các thí nghiệm về đánh giá tác động dòng khí hóa sau đốtmô tả các thí nghiệm về đánh giá tác động dòng khí hóa sau đốt
mô tả các thí nghiệm về đánh giá tác động dòng khí hóa sau đốt
 
ESR spectroscopy in liquid food and beverages.pptx
ESR spectroscopy in liquid food and beverages.pptxESR spectroscopy in liquid food and beverages.pptx
ESR spectroscopy in liquid food and beverages.pptx
 
EWOCS-I: The catalog of X-ray sources in Westerlund 1 from the Extended Weste...
EWOCS-I: The catalog of X-ray sources in Westerlund 1 from the Extended Weste...EWOCS-I: The catalog of X-ray sources in Westerlund 1 from the Extended Weste...
EWOCS-I: The catalog of X-ray sources in Westerlund 1 from the Extended Weste...
 
Comparing Evolved Extractive Text Summary Scores of Bidirectional Encoder Rep...
Comparing Evolved Extractive Text Summary Scores of Bidirectional Encoder Rep...Comparing Evolved Extractive Text Summary Scores of Bidirectional Encoder Rep...
Comparing Evolved Extractive Text Summary Scores of Bidirectional Encoder Rep...
 
Shallowest Oil Discovery of Turkiye.pptx
Shallowest Oil Discovery of Turkiye.pptxShallowest Oil Discovery of Turkiye.pptx
Shallowest Oil Discovery of Turkiye.pptx
 
Bob Reedy - Nitrate in Texas Groundwater.pdf
Bob Reedy - Nitrate in Texas Groundwater.pdfBob Reedy - Nitrate in Texas Groundwater.pdf
Bob Reedy - Nitrate in Texas Groundwater.pdf
 
Topic: SICKLE CELL DISEASE IN CHILDREN-3.pdf
Topic: SICKLE CELL DISEASE IN CHILDREN-3.pdfTopic: SICKLE CELL DISEASE IN CHILDREN-3.pdf
Topic: SICKLE CELL DISEASE IN CHILDREN-3.pdf
 
Nucleophilic Addition of carbonyl compounds.pptx
Nucleophilic Addition of carbonyl  compounds.pptxNucleophilic Addition of carbonyl  compounds.pptx
Nucleophilic Addition of carbonyl compounds.pptx
 
原版制作(carleton毕业证书)卡尔顿大学毕业证硕士文凭原版一模一样
原版制作(carleton毕业证书)卡尔顿大学毕业证硕士文凭原版一模一样原版制作(carleton毕业证书)卡尔顿大学毕业证硕士文凭原版一模一样
原版制作(carleton毕业证书)卡尔顿大学毕业证硕士文凭原版一模一样
 
Applied Science: Thermodynamics, Laws & Methodology.pdf
Applied Science: Thermodynamics, Laws & Methodology.pdfApplied Science: Thermodynamics, Laws & Methodology.pdf
Applied Science: Thermodynamics, Laws & Methodology.pdf
 
bordetella pertussis.................................ppt
bordetella pertussis.................................pptbordetella pertussis.................................ppt
bordetella pertussis.................................ppt
 
Chapter 12 - climate change and the energy crisis
Chapter 12 - climate change and the energy crisisChapter 12 - climate change and the energy crisis
Chapter 12 - climate change and the energy crisis
 
SAR of Medicinal Chemistry 1st by dk.pdf
SAR of Medicinal Chemistry 1st by dk.pdfSAR of Medicinal Chemistry 1st by dk.pdf
SAR of Medicinal Chemistry 1st by dk.pdf
 
Phenomics assisted breeding in crop improvement
Phenomics assisted breeding in crop improvementPhenomics assisted breeding in crop improvement
Phenomics assisted breeding in crop improvement
 
Nucleic Acid-its structural and functional complexity.
Nucleic Acid-its structural and functional complexity.Nucleic Acid-its structural and functional complexity.
Nucleic Acid-its structural and functional complexity.
 

A Proposal of Two-step Autoregressive Model

  • 1. A Proposal of Two-step Autoregressive Model Prof. Dr. Loc Nguyen PhD, PostDoc Loc Nguyen’s Academic Network, Vietnam Email: ng_phloc@yahoo.com Homepage: www.locnguyen.net Two-step Autoregressive Model - Loc Nguyen 14/12/2022 1
  • 2. Abstract Autoregressive (AR) model and conditional autoregressive (CAR) model are specific regressive models in which independent variables and dependent variable imply the same object. They are powerful statistical tools to predict values based on correlation of time domain and space domain, which are useful in epidemiology analysis. In this research, I combine them by the simple way in which AR and CAR is estimated in two separate steps so as to cover time domain and space domain in spatial-temporal data analysis. Moreover, I integrate logistic model into CAR model, which aims to improve competence of autoregressive models. 2 Two-step Autoregressive Model - Loc Nguyen 14/12/2022
  • 3. Table of contents 1. Two-step autoregressive model 2. Integrated with logistic function 3. Conclusions 3 Two-step Autoregressive Model - Loc Nguyen 14/12/2022
  • 4. 1. Two-step autoregressive model Autoregressive (AR) model evaluates new value of a variable based on its old values; in other words, independent variables called regressors and dependent variable called responsor refer to the same concerned object (CO). AR model is often used to predict new values in time series, which is called temporal autoregressive model. Conditional autoregressive (CAR) model also follows ideology of AR model where regressors and responsor imply the same CO but CAR model added more values of neighbors of CO. Therefore, CAR model is often used to predict new values in space domain, which is called spatial autoregressive model. There are some researches like (Mariella & Tarantino, 2010) to integrate AR model and CAR model but here I combine them consequently by the simple way in which AR and CAR is estimated in two separate processes. In other words, the research uses mainly autoregressive model through two steps: 1. Step 1: CAR model known as spatial autoregressive model is estimated to predict responsor based on its neighbors and its environmental features like humidity, precipitation, etc. As a convention, responsor is called case like disease case in epidemiology. 2. Step 2: AR model known as temporal autoregressive model is estimated to forecast next case based on current cases. 14/12/2022 Two-step Autoregressive Model - Loc Nguyen 4
  • 5. 1. Two-step autoregressive model Let S = {1, 2,…, n} be spatial domain and so each site s can correspond to an area. Let nb(s) be set of neighbor sites of site s. Let Xs = (xs1, xs2,…, xsk) be the vector of environmental features at site s and so xsj is the jth environmental feature at site s. Let Ys be the outcome number of cases at site s. If time is concerned, Xs and Ys are replaced by Xst and Yst, respectively. Suppose the time interval is T = 2, this research firstly uses spatial autoregressive model to estimate the number of cases Yst = Ys at current timepoint t. Later on, the research uses temporal autoregressive model to forecast the future number of cases Ys(t+1) at timepoint t+1 based on current Yst. Therefore, the computational model of the research is called the two-step autoregressive model. In the first step, spatial autoregressive model is estimated. Given current timepoint, we have Ys = Yst and Xs = Xst. When building up spatial autoregressive model, we ignore timepoints in dataset. Because the number of cases is moved around an average number, Poisson distribution is more suitable to build spatial autoregressive model. However, for simplicity, I use normal distribution for spatial autoregressive model. Let W be the symmetric weighted adjacency matrix (Mariella & Tarantino, 2010, p. 227): 𝑊 = 𝑤𝑠𝑠∗ where 𝑤𝑠𝑠∗ = 0 if 𝑠∗ = 𝑠 𝜑 𝑠, 𝑠∗ if 𝑠∗ ∈ nb 𝑠 0 otherwise (1.1) Where φ(s, s*) measures the approximation of site s and site s* with note that nb(s) denotes the set of neighbors of site s. Moreover, the set nb(s) does not contain s. 14/12/2022 Two-step Autoregressive Model - Loc Nguyen 5
  • 6. 1. Two-step autoregressive model Because W is symmetric, we have (Mariella & Tarantino, 2010, p. 227): ∀𝑠, 𝑠∗ ∈ 𝑆, 𝑤𝑠𝑠∗ = 𝑤𝑠∗𝑠 Note, W is constant. Let WD be the diagonal matrix of normalization (Mariella & Tarantino, 2010, p. 227): 𝑊𝐷 = diag 𝑤1+, 𝑤2+, … , 𝑤𝑛+ 𝑤𝑠+ = 𝑠∗∈nb 𝑠 𝑤𝑠𝑠∗ (1.2) Let A be the matrix of interaction (Mariella & Tarantino, 2010, p. 227): 𝐴 = 𝑎𝑠𝑠∗ = 𝑊𝐷 −1 𝑊 𝑎𝑠𝑠∗ = 𝑤𝑠𝑠∗ 𝑤𝑠+ (1.3) Note, A is constant. Let ΣD be the diagonal matrix of variances (Mariella & Tarantino, 2010, p. 227): Σ𝐷 = diag 𝜎1 2 , 𝜎2 2 , … , 𝜎𝑛 2 = 𝜎2 𝑊𝐷 −1 𝜎𝑠 2 = 𝜎2 𝑤𝑠+ (1.4) Note, σ2 is called global variance. 14/12/2022 Two-step Autoregressive Model - Loc Nguyen 6
  • 7. 1. Two-step autoregressive model The probability density function (PDF) of the number of cases Ys at Xs given its neighbors Xs* is defined according to normal distribution as follows (Mariella & Tarantino, 2010, p. 227): 𝑓 𝑌𝑠 𝑋𝑠, 𝑋𝑠∗, 𝛼𝑠, 𝜎2 = 1 2𝜋𝜎𝑠 2 exp − 𝑌𝑠 − 𝛼𝑠 𝑇𝑋𝑠 + 𝜌 𝑠∗∈nb 𝑠 𝑎𝑠𝑠∗ 𝑌𝑠∗ − 𝛼𝑠∗ 𝑇 𝑋𝑠∗ 2 2𝜎𝑠 2 (1.5) Where αs = (αs1, αs2,…, αsk) is site regressive coefficient vector and σs 2 is site variance. The constant ρ is called effective neighborhood constant, |ρ| < 1. Each site variance (local variance) σs 2 was defined. 𝜎𝑠 2 = 𝜎2 𝑤𝑠+ Therefore, only parameters αs and σ2 are parameters in the PDF of Ys at Xs given its neighbors. The mean and variance of Ys at Xs given its neighbors are: mean 𝑌𝑠 𝑋𝑠, 𝑋𝑠∗, 𝛼𝑠, 𝜎2 = 𝛼𝑠 𝑇 𝑋𝑠 + 𝜌 𝑠∗∈nb 𝑠 𝑎𝑠𝑠∗ 𝑌𝑠∗ − 𝛼𝑠∗ 𝑇 𝑋𝑠∗ var 𝑌𝑠 𝑋𝑠, 𝑋𝑠∗, 𝛼𝑠, 𝜎2 = 𝜎𝑠 2 (1.6) From the mean of Ys at Xs given its neighbors, the spatial autoregressive model of the number of cases is defined as follows: 𝑌𝑠 = 𝛼𝑠 𝑇𝑋𝑠 + 𝜌 𝑠∗∈nb 𝑠 𝑎𝑠𝑠∗ 𝑌𝑠∗ − 𝛼𝑠∗ 𝑇 𝑋𝑠∗ (1.7) 14/12/2022 Two-step Autoregressive Model - Loc Nguyen 7
  • 8. 1. Two-step autoregressive model The essence of the first step in this research is to estimate all regressive coefficients αs when the effective neighborhood constant ρ is pre-defined. Let xs (i) be the ith instance of the vector of environmental feature Xs at site s. Let ys (i) be the ith instance of Ys at site s. Let Xs be the set of Ns instances xs (i) and let Ys be the set of Ns instances y(i). Each site s owns one dataset (Xs, Ys). 𝑿𝑠 = 𝒙𝑠 1 𝑇 𝒙𝑠 2 𝑇 ⋮ 𝒙𝑠 𝑁𝑠 𝑇 = 1 𝑥𝑠1 1 𝑥𝑠2 1 ⋯ 𝑥𝑠𝑛 1 1 𝑥𝑠1 2 𝑥𝑠2 2 ⋯ 𝑥𝑠𝑛 2 ⋮ ⋮ ⋮ ⋱ ⋮ 1 𝑥𝑠1 𝑁𝑠 𝑥𝑠2 𝑁𝑠 ⋯ 𝑥𝑠𝑛 𝑁𝑠 𝒙𝑠 𝑖 = 1 𝑥𝑠1 𝑖 𝑥𝑠2 𝑖 ⋮ 𝑥𝑠𝑛 𝑖 , 𝒙𝑠 𝑗 = 1 𝑥𝑠1 𝑗 𝑥𝑠2 𝑗 ⋮ 𝑥𝑠𝑛 𝑗 , 𝒚𝑠 = 𝑦𝑠 1 𝑦𝑠 2 ⋮ 𝑦𝑠 𝑁𝑠 (1.8) 14/12/2022 Two-step Autoregressive Model - Loc Nguyen 8
  • 9. 1. Two-step autoregressive model Let X and Y be the sets of all site dataset Xs and Ys, respectively. Let α and σ2 be the sets of all αs and all σs 2, respectively. The joint probability of the global dataset (X, Y) is: 𝑓 𝑿, 𝒀 𝛼, 𝜎2 = 𝑠=1 𝑆 𝑓 𝑌𝑠 𝑋𝑠, 𝑋𝑠∗, 𝛼𝑠, 𝜎𝑠 2 = 𝑠=1 𝑆 2𝜋𝜎𝑠 2 − 𝑁𝑠 2 ∗ 𝑠=1 𝑆 exp − 1 2𝜎𝑠 2 𝑖=1 𝑁𝑠 𝑦𝑠 𝑖 − 𝛼𝑠 𝑇𝒙𝑠 𝑖 + 𝜌 𝑠∗∈nb 𝑠 𝑎𝑠𝑠∗ 𝑦𝑠∗ 𝑖 − 𝛼𝑠∗ 𝑇 𝒙𝑠∗ 𝑖 2 The log-likelihood function is logarithm function of such joint probability: 𝑙 𝛼, 𝜎2 = − 1 2 𝑠=1 𝑆 𝑁𝑠log 2𝜋𝜎𝑠 2 − 1 2 𝑠=1 𝑆 1 𝜎𝑠 2 𝑖=1 𝑁𝑠 𝑦𝑠 𝑖 − 𝛼𝑠 𝑇 𝒙𝑠 𝑖 + 𝜌 𝑠∗∈nb 𝑠 𝑎𝑠𝑠∗ 𝑦𝑠∗ 𝑖 − 𝛼𝑠∗ 𝑇 𝒙𝑠∗ 𝑖 2 Let αs * = (αs1 *, αs2 *,…, αsk *) be the optimal estimate of regressive coefficients and thus, αs * is a maximizer of l(α, σ2). 𝛼𝑠 ∗ = argmax 𝛼1,𝛼2,…,𝛼𝑠,…,𝛼 𝑆 𝑙 𝛼, 𝜎2 14/12/2022 Two-step Autoregressive Model - Loc Nguyen 9
  • 10. 1. Two-step autoregressive model By taking first-order partial derivatives of l(α, σ2) with regard to αs, we obtain: 𝜕𝑙 𝛼, 𝜎2 𝜕𝛼𝑠 = 1 𝜎𝑠 2 𝑖=1 𝑁𝑠 𝑦𝑠 𝑖 − 𝛼𝑠 𝑇𝒙𝑠 𝑖 + 𝜌 𝑠∗∈nb 𝑠 𝑎𝑠𝑠∗ 𝑦𝑠∗ 𝑖 − 𝛼𝑠∗ 𝑇 𝒙𝑠∗ 𝑖 𝒙𝑠 𝑖 𝑇 When the first-order partial derivatives of l(α, σ2) with regard to αs are equal to zero, it gets locally maximal. In other words, αs * is solution of the following equation resulted from setting such derivatives to be zero. 𝑖=1 𝑁𝑠 𝑦𝑠 𝑖 − 𝛼𝑠 𝑇𝒙𝑠 𝑖 + 𝜌 𝑠∗∈nb 𝑠 𝑎𝑠𝑠∗ 𝑦𝑠∗ 𝑖 − 𝛼𝑠∗ 𝑇 𝒙𝑠∗ 𝑖 𝒙𝑠 𝑖 𝑇 = 𝟎𝑇 As a convention, ass* = 0 if ys* (i) or xs* (i) does not exist. The equation above is re-written as follows: 𝒚𝑠 𝑇 𝑿𝑠 − 𝛼𝑠 𝑇 𝑿𝑠 𝑇 𝑿𝑠 − 𝜌 𝑠∗∈nb 𝑠 𝑎𝑠𝑠∗ 𝒚𝑠∗ 𝑇 𝑿𝑠 − 𝛼𝑠∗ 𝑇 𝑿𝑠 𝑇 𝑿𝑠 = 𝒚𝑠 𝑇 − 𝜌 𝑠∗∈nb 𝑠 𝑎𝑠𝑠∗𝒚𝑠∗ 𝑇 𝑿𝑠 − 𝛼𝑠 𝑇 − 𝜌 𝑠∗∈nb 𝑠 𝑎𝑠𝑠∗𝛼𝑠∗ 𝑇 𝑿𝑠 𝑇𝑿𝑠 = 𝟎𝑇 This implies 𝛼𝑠 − 𝜌 𝑠∗∈nb 𝑠 𝑎𝑠𝑠∗𝛼𝑠∗ = 𝑿𝑠 𝑇𝑿𝑠 −1𝑿𝑠 𝑇 𝒚𝑠 − 𝜌 𝑠∗∈nb 𝑠 𝑎𝑠𝑠∗𝒚𝑠∗ 14/12/2022 Two-step Autoregressive Model - Loc Nguyen 10
  • 11. 1. Two-step autoregressive model In general, the estimates αs * for all regressive coefficient vectors αs where s from 1 to |S| is solution of the linear equation system consisting of |S| equations as follows: 𝛼𝑠 − 𝜌 𝑠∗∈𝑛𝑏 𝑠 𝑎𝑠𝑠∗𝛼𝑠∗ = 𝑿𝑠 𝑇 𝑿𝑠 −1 𝑿𝑠 𝑇 𝒚𝑠 − 𝜌 𝑠∗∈𝑛𝑏 𝑠 𝑎𝑠𝑠∗𝒚𝑠∗ 𝑠=1, 𝑆 (1.9) It is not difficult to solve the linear equation system above. As a result, the spatial autoregressive model is re-written as follows: 𝑌𝑠 = 𝛼𝑠 ∗ 𝑇 𝑋𝑠 + 𝜌 𝑠∗∈nb 𝑠 𝑎𝑠𝑠∗ 𝑌𝑠∗ − 𝛼𝑠∗ ∗ 𝑇 𝑋𝑠∗ (1.10) 14/12/2022 Two-step Autoregressive Model - Loc Nguyen 11
  • 12. 1. Two-step autoregressive model In the second step, temporal autoregressive model is estimated. For simplicity, we follow the Markov condition in which, future cases only depend on current cases. When building up spatial autoregressive model, we ignore environmental features Xs in dataset or we focus on a given site s. Suppose the current number of cases Yst at site s was determined previously by spatial autoregressive model. As a convention, the “s” index is removed from the response variable and so we have Yt+1 = Ys(t+1) and Yt = Yst. We will forecast next (unknown) Yt+1 based on current (known) Yt. The temporal autoregressive model is defined as follows (Eshel, 2003, p. 1): 𝑌𝑡+1 = 𝛽𝑌𝑡 + 𝜀 (1.11) Where ε is the error random variable and β is temporal coefficient with note that mean of ε is 0 such that 𝐸 𝑌𝑡+1 = 𝛽𝑌𝑡 Let yt be the value of Yt at timepoint t. Suppose there are T timepoints and we have t = 1, 2,…, T. The sum of squared errors is defined as follows: 𝑆 𝛽 = 𝑡=1 𝑇−1 𝑦𝑡+1 − 𝛽𝑦𝑡 2 Let β* be the estimate of β. The least squares method sets β* as the minimizer of S(β). 𝛽∗ = argmin 𝛽 𝑆 𝛽 14/12/2022 Two-step Autoregressive Model - Loc Nguyen 12
  • 13. 1. Two-step autoregressive model By taking first-order derivative of S(β) with regard to β, we obtain (Eshel, 2003, p. 1): 𝑑𝑆 𝛽 𝑑𝛽 = −2 𝑡=1 𝑇−1 𝑦𝑡+1 − 𝛽𝑦𝑡 𝑦𝑡 The temporal coefficient β* is solution of the following equation resulted from setting such derivative to be zero. −2 𝑡=1 𝑇−1 𝑦𝑡+1 − 𝛽𝑦𝑡 𝑦𝑡 = 0 It is easy to determine the formula for calculating β* (Eshel, 2003, p. 1). 𝛽∗ = 𝑡=1 𝑇−1 𝑦𝑡𝑦𝑡+1 𝑡=1 𝑇−1 𝑦𝑡 2 (1.12) Please pay attention that values yt are calculated by CAR model estimated in the first step. The temporal autoregressive model is re-written as follows: 𝑌𝑠 𝑡+1 = 𝛽∗𝑌𝑠𝑡 (1.13) Equation 1.12 to calculate β* is simplest form of the well-known Yule-Walker equation. 14/12/2022 Two-step Autoregressive Model - Loc Nguyen 13
  • 14. 2. Integrated with logistic function In normal CAR model, the responsor Ys which is the outcome number of cases at site s is real variable but in the situation that Ys is category variable whose G ordered values are v1, v2,…, vG, new modification of CAR model is required. The multinomial logistic function of Ys with regard to value vg at site s, which is associated spatial autoregression, is proposed as follows: 𝑓𝑔 𝑌𝑠 𝑋𝑠, 𝑋𝑠∗, 𝛼𝑠𝑔 = 1 1 + exp − 𝛼𝑠𝑔 𝑇 𝑋𝑠 + 𝜌 𝑠∗∈nb 𝑠 𝑎𝑠𝑠∗ 𝛼𝑠𝑔 𝑇 𝑋𝑠∗ − 𝛼𝑠∗𝑔 𝑇 𝑋𝑠∗ (2.1) Note that αsg = (αsg1, αsg2,…, αsgk) which is site regressive coefficient vector with regard to value yg is the main parameter which is estimated here. Let pg be the probability that Ys receives value yg, where g is from 1 to G. If G values v1, v2,…, vG are not ordered, pg is the same to the proposed multinomial logistic function of Ys. 𝑝 𝑋𝑠 𝛼𝑠𝑔 = 𝑃 𝑌𝑠 = 𝑣𝑔 𝑋𝑠, 𝑋𝑠∗, 𝛼𝑠𝑔 = 𝑓𝑔 𝑌𝑠 𝑋𝑠, 𝑋𝑠∗, 𝛼𝑠𝑔 = 1 1 + exp − 𝛼𝑠𝑔 𝑇 𝑋𝑠 + 𝜌 𝑠∗∈nb 𝑠 𝑎𝑠𝑠∗ 𝛼𝑠𝑔 𝑇 𝑋𝑠∗ − 𝛼𝑠∗𝑔 𝑇 𝑋𝑠∗ (2.2) If G values v1, v2,…, vG are ordered such that v1 ≤ v2 ≤ … ≤ vG then, according to Bender and Grouven (Bender & Grouven, 1997, p. 547), pg becomes the cumulative probability such that Ys ≤ vg as follows: 𝑝 𝑋𝑠 𝛼𝑠𝑔 = 𝑃 𝑌𝑠 ≤ 𝑣𝑔 𝑋𝑠, 𝑋𝑠∗, 𝛼𝑠𝑔 = 𝑖=1 𝑔 𝑓𝑖 𝑌𝑠 𝑋𝑠, 𝑋𝑠∗, 𝛼𝑠𝑖 (2.3) Obviously, p(Xs|αsg) is the CAR model integrated with logistic function, which is called LCAR model. 14/12/2022 Two-step Autoregressive Model - Loc Nguyen 14
  • 15. 2. Integrated with logistic function Essentially, LCAR is determined by estimating parameters αsg. As a convention, let 𝑝𝑠𝑔 = 𝑝 𝑋𝑠 𝛼𝑠𝑔 𝑔=1 𝐺 𝑝𝑠𝑔 = 1 (2.4) The logarithm of psg which is called logit of psg is defined as follows: logit 𝑝𝑠𝑔 = log 𝑝𝑠𝑔 1 − 𝑝𝑠𝑔 = 𝛼𝑠𝑔 𝑇 𝑋𝑠 + 𝜌 𝑠∗∈nb 𝑠 𝑎𝑠𝑠∗ 𝛼𝑠𝑔 𝑇 𝑋𝑠∗ − 𝛼𝑠∗𝑔 𝑇 𝑋𝑠∗ (2.5) Suppose Ys obeys multinomial distribution so that the parameters αsg at site s are determined by maximum likelihood estimation (MLE) method (Czepiel, 2002). The joint probability of the global dataset (X, Y) in reduced form is: 𝑓 𝑿, 𝒀 𝛼 = 𝑠=1 𝑆 𝑔=1 𝐺 𝑖=1,𝑦𝑠 𝑖 =𝑣𝑔 𝑁𝑠 𝑝𝑠𝑔 𝑖 Where psg (i) is the evaluation of psg with the instance Xs = xs (i). The log-likelihood function is logarithm function of such joint probability: 𝑙 𝛼 = − 𝑠=1 𝑆 𝑔=1 𝐺 𝑖=1,𝑦𝑠 𝑖 =𝑣𝑔 𝑁𝑠 log 1 + exp − 𝛼𝑠𝑔 𝑇 𝒙𝑠 𝑖 + 𝜌 𝑠∗∈nb 𝑠 𝑎𝑠𝑠∗ 𝛼𝑠𝑔 𝑇 𝒙𝑠∗ 𝑖 − 𝛼𝑠∗𝑔 𝑇 𝒙𝑠∗ 𝑖 14/12/2022 Two-step Autoregressive Model - Loc Nguyen 15
  • 16. 2. Integrated with logistic function Let αsg * = (αsg1 *, αsg2 *,…, αsgk *) be the optimal estimate of regressive coefficients which is a maximizer of l(α) with the constraint 𝑔=1 𝐺 𝑝𝑠𝑔 = 1 which is equivalent to: log 𝑔=1 𝐺 𝑝𝑠𝑔 = 0 Let L0(α, λ) be Lagrange function defined as follows: 𝐿0 𝛼, 𝜆 = 𝑙 𝛼 + 𝑠=1 𝑆 𝑔=1 𝐺 𝑖=1,𝑦𝑠 𝑖 =𝑣𝑔 𝑁𝑠 𝜆𝑖 𝑠𝑔 log 𝑝𝑠𝑔 𝑖 Where λi are Lagrange multipliers such that λi ≥ 0 with note that each λi (sg) implies λi at site s given value vg. Because logarithm is concave function, we apply Jessen inequation to approximate the Lagrange function as follows: 𝐿0 𝛼, 𝜆 ≤ 𝐿 𝛼, 𝜆 = 𝑙 𝛼 + 𝑠=1 𝑆 𝑔=1 𝐺 𝑖=1,𝑦𝑠 𝑖 =𝑣𝑔 𝑁𝑠 𝜆𝑖 𝑠𝑔 log 𝑝𝑠𝑔 𝑖 We now focus on maximizing L(α, λ). 14/12/2022 Two-step Autoregressive Model - Loc Nguyen 16
  • 17. 2. Integrated with logistic function We now focus on maximizing L(α, λ). The first-order partial derivative of L(α, λ) with regard to αsg is: 𝜕𝐿 𝛼, 𝜆 𝜕𝛼𝑠𝑔 = 𝑖=1,𝑦𝑠 𝑖 =𝑣𝑔 𝑁𝑠 𝜆𝑖 + 1 1 − 𝑝𝑠𝑔 𝑖 𝒙𝑠 𝑖 𝑇 + 𝜌 𝑠∗∈nb 𝑠 𝑎𝑠𝑠∗ 𝒙𝑠∗ 𝑖 𝑇 Note, ass* = 0 if ys* (i) or xs* (i) does not exist. Let 𝑍𝑠 𝑖 = 𝒙𝑠 𝑖 + 𝜌 𝑠∗∈nb 𝑠 𝑎𝑠𝑠∗𝒙𝑠∗ 𝑖 (2.6) We have: 𝜕𝐿 𝛼, 𝜆 𝜕𝛼𝑠𝑔 = 𝑖=1,𝑦𝑠 𝑖 =𝑣𝑔 𝑁𝑠 𝜆𝑖 + 1 1 − 𝑝𝑠𝑔 𝑖 𝑍𝑠 𝑖 𝑇 Setting the first-order partial derivative of L(α, λ) with regard to λi to be 0, we obtain: 𝜕𝐿 𝛼, 𝜆 𝜕𝜆𝑖 = 𝜆𝑖 𝑔=1 𝐺 log 𝑝𝑠𝑔 𝑖 = 0 ⇒ 𝜆𝑖 = 0 Substituting all λi = 0 into the derivative 𝜕𝐿 𝛼,𝜆 𝜕𝛼𝑠𝑔 , we have: 𝜕𝐿 𝛼, 𝜆 𝜕𝛼𝑠𝑔 = 𝑖=1,𝑦𝑠 𝑖 =𝑣𝑔 𝑁𝑠 1 − 𝑝𝑠𝑔 𝑖 𝑍𝑠 𝑖 𝑇 14/12/2022 Two-step Autoregressive Model - Loc Nguyen 17
  • 18. 2. Integrated with logistic function By setting such derivative 𝜕𝐿 𝛼,𝜆 𝜕𝛼𝑠𝑔 to be zero, we obtain: 𝑖=1,𝑦𝑠 𝑖 =𝑣𝑔 𝑁𝑠 𝑝𝑠𝑔 𝑖 𝑍𝑠 𝑖 = 𝑖=1,𝑦𝑠 𝑖 =𝑣𝑔 𝑁𝑠 𝑍𝑠 𝑖 As a result, the estimates αsg * for all regressive coefficient vectors αsg over all sites and all values v1, v2,…, vG is solution of the linear equation system consisting of |S|G equations as follows: 𝑖=1,𝑦𝑠 𝑖 =𝑣𝑔 𝑁𝑠 𝑝𝑠𝑔 𝑖 𝑍𝑠 𝑖 = 𝑖=1,𝑦𝑠 𝑖 =𝑣𝑔 𝑁𝑠 𝑍𝑠 𝑖 𝑠=1, 𝑆 ,𝑔=1,𝐺 (2.7) In general, the equation 2.7 is used to estimate LCAR model, which can be solved by numerical methods like Newton-Raphson method for nonlinear equation system (Burden & Faires, 2011, pp. 638-642). 14/12/2022 Two-step Autoregressive Model - Loc Nguyen 18
  • 19. 3. Conclusions In this research, I proposed a simple integration model for spatial- temporal autoregression which consists of two consequent and separated steps. Moreover, I also proposed the integration of logistic model and autoregressive model called LCAR which is slightly different from traditional CAR model. However, the traditional AR model known as temporal autoregressive model is still simple with assumption of first-order Markov property. In the future trend, I will improve AR model with autoregressive moving average (ARMA) and autoregressive integrated moving average (ARIMA). 14/12/2022 Two-step Autoregressive Model - Loc Nguyen 19
  • 20. References 1. Bender, R., & Grouven, U. (1997, September 5). Ordinal logistic regression in medical research. Journal of the Royal College of Physicians of London, 31(5), 546-551. Retrieved from https://pubmed.ncbi.nlm.nih.gov/9429194/ 2. Burden, R. L., & Faires, D. J. (2011). Numerical Analysis (9th Edition ed.). (M. Julet, Ed.) Brooks/Cole Cengage Learning. 3. Czepiel, S. A. (2002). Maximum Likelihood Estimation of Logistic Regression Models: Theory and Implementation. Czepiel's website http://czep.net. 4. Eshel, G. (2003). The Yule Walker Equations for the AR Coefficients. University of Wharton, Statistics Department. Michael Steele Homepage. Retrieved 11 24, 2018, from http://www- stat.wharton.upenn.edu/~steele/Courses/956/ResourceDetails/YWSourceFiles/YW- Eshel.pdf 5. Mariella, L., & Tarantino, M. (2010). Spatial Temporal Conditional Auto-Regressive Model: A New Autoregressive Matrix. Austrian Journal of Statistics, 39(3), 223-244. doi:10.17713/ajs.v39i3.246 14/12/2022 Two-step Autoregressive Model - Loc Nguyen 20
  • 21. Thank you for listening 21 Two-step Autoregressive Model - Loc Nguyen 14/12/2022