Download as pdf or txt
Download as pdf or txt
You are on page 1of 28

Analytical Framework for

Adoption Studies

Moises Neil V. Seriño, PhD


Director, Visayas Socio-Economic Research & Data Analytics Center
Dean, College of Management and Economics
Visayas State University
Adoption
Productivity & Quantification
Study
Efficiency & Qualification
Analysis of Impacts
Productivity
& Efficiency
Analysis

Adoption
Studies

Quantification
&
Qualification
of Impacts
LIMDEP Models
A. Dummy Dependent Variable
i. Linear probability model (LPM)
ii. Probit model
iii. Logit model
iv. Multinomial Logit / Ordered Logit
B. Censored Dependent Variable
i. Tobit model
LIMDEP Models

• LIMDEP – Limited Dependent Variable


Model
Factors affecting adoption

• Social factors: • Physical factors:


• Age of farmer • Farm area
• Education • Topography
• Experience in farming • Soil Type
• Cropping intensity

• Institutional factors: • Economic factors:


• Presence of subsidy • Family income
• Gov’t Program • Opportunity cost
• Awareness campaign • Risk of adoption
• Training availablity • Credit
Example: Adoption of a new
technology

Adoption = f(Age, FarmSize, Education …)


Adoption = 1 if farmer is an adopter
=0 if farmer is not an adopter
Note: The dependent variable – dichotomous in nature
e.g. Yes - No categories

%Adoption = f(Credit, Training, Subsidy…….)


%Adoption = 1 if the farmer is a full adopter
= 2 if the farmer is a partial adopter
= 3 if the farmer is not an adopter

Note: The dependent variable has more than 2 categories


A. LPM – Linear Probability Model

Problems of LPM
1. Predicted probability P^from LPM regression is
not guaranteed to fall in the range of 0-1.
2. Marginal probability per unit change in X is
constant and is equal to the slope Bi.

This is an unrealistic assumption! Why?


A. LPM – Linear Probability Model
^

* actual observations are “bunched up” along P = 0 and P = 1


- predicted values can lie outside (0,1)
- prediction errors can be very large

Note: The adoption


variable has only
two values, i.e.,
0 and 1
Example

Adopti = β0 + β1Age + β2Farm + β3Educ +µi

Adopti = 1 if farmer is an adopter


= 0 otherwise
Age - age of the farmer
Farm - Size of the farm

Educ = 1 if the farmer has finished beyond


elementary (either HS or College)
= 0 if the farmer finished only
elementary schooling
A. LPM Regression Result

Note: The dependent variable is interpreted as probability.

P(Adopt)= 2.4980 - 0.0464Age + 0.011Farm + 0.3786Educ


R2 = 0.4159
B1 = -0.0464 if age of the farmer increases by one
year, probability of adopting the new
technology decreases by 0.0464 or
4.64%
B3 = 0.3786 if the farmer had more than high school
education, the probability of adopting the
new technology increases by 0.3786 or
37.86%
A. LPM Estimates of Probability

Farmer #1 Farmer #20 Farmer #47


Age =50 years Age =34 years Age = 70 years
Farm=1 ha Farm=2 has Farm=0.5 ha
Educ=0 Educ=1 Educ=0

P(Adopt) =0.189 P(Adopt) =1.321 P(Adopt) = -0.774

The probability that farmer #1


will adopt the new technology Probability Negative
is 18.9% greater than 1! Probability!

Therefore, LPM is not recommended for LIMDEP


relationships!
B. Probit Model
^
to solve the problems of LPM, need transformation so that P
will lie in the interval 0-1

* The cumulative normal function will be a suitable


distribution to capture the behavior of probability of
adoption

Note:
The probability
function tends to
exhibit a double
threshold equal to 0
and 1
B. The Probability Distribution

Pi = F(Zi)
where Pi – probability
F -- cumulative density function
Zi = βo + β1X + µ Regression model
with Z as dependent
variable

NOTE: Z is the standard normal distribution with mean 0


and variance equal to 1
Using the standardized Z table, the probability of
adoption P[Zc ≤ Ztab] can be computed.
B. The Probability Distribution
The probability of adoption can also be computed using the
cumulative normal formula:
Z2
1 Ζι - Note:
Pi = F(Zi ) =

∫∞ e 2
dZ π = 3.1416

where Z – is the predicted value from the probit regression


given the factors that affect adoption.

Cumulative
normal
distribution
B. Probit Regression Result

Zi = 1.831 – 0.045Age + 0.151Farm + 0.816Educ


pseudo R2 = 0.6593
B1 = -0.045 if age of the farmer increases by one
year, the Z-value of adopting the new
technology decreases by 0.045
B3 = 0.816 if the farmer had more than high school
education, the Z-value of adopting the
new technology increases by 0.816
B. Estimating Probability in Probit
Model
Z = 1.831 – 0.045 Age + 0.151 Farm + 0.816 Educ
Farmer #20 Z =1.831 – 0.045 (34) + 0.151 (2) + 0.816 (1) = 1.419
Farmer #20 Farmer #47
Age = 34 Age = 70
Farm = 2 Farm = 0.5
Educ =1 Educ =0
Z=1.419 Z=-1.2435
Probability of Adoption =0.1462 Probability of Adoption =0.184662

Solving for the cumulative normal density function


1.419 -1.2435
1 Z2
1 Z2
Pi = F(Zi) =

∫e -
2 dZ Pi = F(Zi) =

∫ e - 2 dZ
- ∞ - ∞

1 ⎡ (1.419)2 ⎤ 1 ⎡ (-1.2435)2 ⎤
Pi = F(Zi) = e⎢ -2
⎥⎦ Pi = F(Zi) = e⎢ -2
⎥⎦
2(3.1416) ⎣ 2(3.1416) ⎣
= 0.40 (0.3654) = 0.40 (0.4616)
= 0.1462 P is now less than 1.0 = 0.1846 P is now positive
C. Logit Model
Adopti = β0 + β1Age + β2Farm + β3Educ + µi
assumed to be cumulative
normal in probit model

The cumulative logistic distribution is used as the


underlying distribution for the model

Logit model
C. Logistic Distribution

Pi =
1+ e−z

Pi
Zi = ln
1-Pi

Probit Model: Zi = β0 + β1Age + β2Farm +β3Educ + µi

Pi
Logit Model: ln = β0 + β1Age + β2Farm +β3Educ + µi
1-Pi

where Pi - probability of an event A


(1-Pi ) - probability of event A not happening
C. Logistic Distribution

P= ⎛ P ⎞
1 + e-Z Z = ln ⎜ ⎟
⎝1 – P ⎠
Proof:
1 1 -P
P= e -Z
=
1+ e -Z p

P(1 + e-Z ) = 1 Z P
e =
1 -P

P + Pe-Z = 1 1
⎛ P ⎞
Z lne = ln ⎜ ⎟
⎝1 - P ⎠
Pe -Z = 1 -P
P ⎞
Z = ln ⎛⎜ ⎟ Q.E.D.
⎝1 - P ⎠
C. Logit Regression Result

The dependent variable is interpreted as the “log of the odds


ratio”
Pi
ln = 0.3021 - 0.0231Age + 0.3852 Farm + 0.1784 Educ
1 - Pi
pseudo R2 = 0.6593
B1 = -0.0231 if age of the farmer increases by one
year, the log of the odds ratio of adopting
the new technology decreases by 0.0231
B3 = 0.1784 if the farmer had more than high school
education, the log of the odds ratio of
adopting the new technology increases by
0.1784
C. Logit Probability Estimation
Example1:
Pi
ln = 0.3021 - 0.0231Age+ 0.3852 Farm + 0.1784 Educ
1 -Pi
Pi
ln = 0.3021 - 0.0231(34)+ 0.3852 (2)+ 0.1784 (1)
1 -Pi
Pi
ln = Z = 0.4655
1 -Pi
Farmer #20 Pi
ln = Z =0.4655
1 – Pi
Age = 34
Farm = 2 1
Pi =
Educ =1 1 +e–Z
Z=0.4655 1
Probability of Adoption =0.6143 P i = = 0.6143
1 +e –(0.4655)

Note: The dependent variable is interpreted as log of the odds ratio.


The probability of adoption can be computed using the
cumulative logistic formula.
C. Logit Probability Estimation
Example2: Pi
ln = 0.3021 - 0.0231Age + 0.3852 Farm + 0.1784 Educ
1 - Pi
Pi
ln = 0.3021 - 0.0231(70)+ 0.3852 (0.5)+ 0.1784 (0)
1 - Pi
Pi
ln = Z = -1.1223
1 - Pi
Farmer #47 Pi
ln = Z =–1.1223
– Pi
Age = 70
Farm = 0.5 1
Pi =
Educ =0 1 +e–Z
Z= -1.1223 1
Probability of Adoption =0.2456 Pi = =0.2456
1 +e–(–1.1223)
Rule: The higher the value of the “log of the odds ratio”, the higher
is the odds in favor of adoption. Therefore, the higher is the
probability of adopting the new technology.
C. Logit Probability Estimation
Example2: Pi
ln = 0.3021 - 0.0231Age + 0.3852 Farm + 0.1784 Educ
1 - Pi
Pi
ln = 0.3021 - 0.0231(70)+ 0.3852 (0.5)+ 0.1784 (0)
1 - Pi
Pi
ln = Z = -1.1223
1 - Pi
Farmer #47 Pi
ln = Z =–1.1223
– Pi
Age = 70
Farm = 0.5 1
Pi =
Educ =0 1 +e–Z
Z= -1.1223 1
Probability of Adoption =0.2456 Pi = =0.2456
1 +e–(–1.1223)
Rule: The higher the value of the “log of the odds ratio”, the higher
is the odds in favor of adoption. Therefore, the higher is the
probability of adopting the new technology.
Summary of the 3 LimDep Models
Linear Probability Model (LPM)
Pi = β0 + β1X1 + β2X2 + … + βnXn + µi

where Pi is the probability of the event

Probit Model
Zi = β0 + β1X1 + β2X2 + … +βnXn + µi
where Zi is the standardized normal variable

Logit Model
⎡ Pi ⎤
ln
⎢⎣ 1 - Pi ⎥⎦ = β0+ β1X + β2X 2 + … +βnX n + iµ
1

⎡lnP i ⎤
Where ⎢⎣ is the log of the odds ratio or “logit”
1 - P i ⎥⎦
Hands on LimDeP Model
Workshop
Determinants of Adoption

1. Use adopt2
2. Conduct a descriptive analysis
3. Determine the factors affecting adoption
using OLS, probit, and logit
4. What are the significant variables affecting
adoption
5. Interpret the coefficients of the significant
variables affecting adoption
Thank You!!!

You might also like