Quantitative Analysis

Thursday, 4 January 2024 3:01 pm

• Economic Analysis is used for theory testing, forecasting, or problem solving. It can be either
descriptive or inferential. There are five steps of EA:
1. Formulation
2. Economic or Econometric Model (OLS is default to test only Null Hypothesis i.e. X will not
impact Y)
3. Data Collection
4. Data Analysis
5. Results
• Statistics is based on measure of central tendency or dispersion
• Two tail test are used when +/- of relation is unknown
• Types of Data:
○ Cross Sectional Data: One time collection, sequence not important, multiple respondent, no time
dependence
○ Time Series: Multi time collection, sequence is important, one respondent, time dependence
R1 2021 10%
R2 2022 9%
R3 2023 5%
R4 2024 18%
○ Panel Data/Longitudinal Data/Pooled Cross Sectional Data: Time dependent assumption violated
R1 2021 10%
R1 2022 9%
R2 2021 5%
R2 2022 18%
○ Time Lag Studies: take different variable from the same respondent at different time. No time
interval is important
• Data Distribution:
○ Discrete (finite) - Binominal (that a value will take one of two independent values) & Poisson
(probability of a given number of events occurring in a fixed interval of time)
○ Continuous (infinite) - Gaussian (assumed that during any measurement values will follow a
normal distribution with an equal number of measurements above and below the mean value)
○ Skewness: Positive Skewness = mean>median>mode, Negative Skewness = mode>median>mean
○ Kurtosis is a statistical measure that quantifies the shape of a probability distribution. Positive
kurtosis indicates heavier tails and a more peaked distribution, while negative kurtosis suggests
lighter tails and a flatter distribution. Kurtosis helps in analyzing the characteristics and outliers of
a dataset.
• Regression:
○ Prediction Technique, causal and directional relation can be checked
○ Most basic form of regression is LEAST SQUARE. Square is a measurement of dispersion .
Averages of all dispersions should be zero. The reason why error terms are not used in estimated
equations is in OLS error term=0
▪ Ordinary Least Square (if all assumptions ok)
▪ Generalized Least Square (Use when homoscedasticity assumption violated)
▪ Weighted Least Square (Use when homoscedasticity & autocorrelation assumption violated)
▪ For multiple dependent variables, Generalized Linear Mixed Models (GLMM) is used
▪ Partial Least Square (when assumption of normality is violated)
○ F-Stat tells you about goodness of fit. You cannot proceed further if F-test is not ok. F-Test is also
called ANOVA Value (p-value should be less than 1%). Higher the F-stat, the better.
○ Assumptions of Regression
1. Linearity
□ The relationship between parameters is linear

Quantitative Analysis Page 1

 □ The relationship between parameters is linear
□ Checked through scatter plot
2. Zero-Error Mean
□ Average of predicted residual values should be zero
3. Homoscedasticity (homo=same, scedasti=variance)
□ When the variance of the error terms (e) appears constant over a range of predictor
variables, the data are said to be homoscedastic. For different values x, residual
variance of y should not change.
□ Checked through scatter plot of predicted values of y and error terms of y
4. Multicollinearity
□ In a multivariate model, all independent variables should not be correlated.
□ Checked through Variance Inflation Factor (VIF). Should not be greater than 5 or 10
□ Exclude trouble making variables
If assumption of Homo & Multi are met, then the data is said to have Independent & Identically
Distributed Errors (iid)
5. Autocorrelation
□ It is mostly noted in time series analysis. It means that that the independent variable
has a lower impact than the error term.
□ Use line graph of residuals or Use DW test (0 to 4 range, if 2, then no autocorrelation)
□ If autocorrelation is noted, user lag variable
6. Endogeneity
□ If error terms and x are correlated, then there is an issue of endogeneity. Reverse
causality can be a reason for endogeneity.
□ Checked through Durban-Watson-Hausman Specification Test
□ Run Y=a+Bx+E
□ Save unstandardized predicted values of Yp & Es
□ Run X=a+B1Yp+B2Es+E (if B2 is significant, then there is an issue of endogeneity)
7. Normality of Errors & DV (optional)
□ Produce better results, increase reliability of results
□ Normal probability plot, Q-Q Plot
• Ordinal vs Interval Scale Debate
○ In ordinal scale, the order must be preserved and only statistical median is permissible in the
strictest sense. The people of argue against treating ordinal scale as interval scale have following
issues:
○ There are no equal intervals between two points and non-normality
○ The scales of Y and X should be congruent
○ The solution offered to treat ordinal as interval scale: Phrase Completion Scale (PCS) which has
11 point system because it uses "the use of integers, which are by definition equally spaced"
• Formative vs Reflective Models
Sr. No. Questions Formative Reflective
1.0 Direction of causality Items to construct Construct to items
1.1 Do items define or manifest the construct? Define Manifest
1.2 Would change in items change the construct? Yes No
1.3 Would change in construct change the item? No Yes
2.0 Can items & indicators be used interchangeably? No Yes
2.1 Should items/indicators have similar/same content? No Yes
2.2 Should items/indicators have a common theme? No Yes
3.0 Covariation among items No Yes
3.1 Does change in one item impact the other? No Yes
4.0 Nomological Net May differ Cannot differ
4.1 Antecedents & consequences of items Can be different Should be same
• Ten Steps in Developing a Quantitative Model:
1. Identify DVs

Quantitative Analysis Page 2

 1. Identify DVs
2. Rationale for studying DVs
3. RQ Develop
4. Theory
5. IVs
6. Mediators
7. Moderators
8. Control Variables
9. Identify Population
10. Develop CF
• Survey Design
○ Decide b/w Reflective vs Formative Models (AMOS only reflective)
○ Items should move together
○ Same number of items for each dimension, 4-6 items per dimension
○ Proper use of scale
○ No dichotomous questions
○ Left to right - negative to positive
○ Avoid buckets
○ Avoid N/A, Others
○ Reverse coded items
○ One thing per item to be measures
○ Don't make assumptions, Applicability across population
○ No sensitive questions
○ Attention trap
○ Put DV in beginning
○ Common method bias scale
• Model Testing
• Develop a good theoretical model
○ See the 10 steps again of formulating a decent quantitative model
○ Develop hypotheses to represent your model
• Case Screening
○ Missing data in rows (countblank in excel, frequency table is SPSS)
○ Unengaged responses (SD=0)
○ Outliers (scatter plot of ID & variable)
• Variable Screening
○ Missing data in columns (same as in rows)
○ Skewness (+/-1) & Kurtosis (+/-3_
• Exploratory Factor Analysis
○ Generate clean pattern matrix
○ How to choose between FA & PCA? whether you want to generalize the findings from your
sample to a population, and whether you are exploring your data or testing a specific hypothesis
○ Principal Component Analysis or Principal Axis Factoring or Image Factoring:
assume that the sample used is the population and results cannot be extrapolated beyond that
sample. If you want to generalize the results then you would need to cross-validate the factor
structure in a different sample
○ Maximum Likelihood Method or Kaiser’s Alpha Factoring: assumes that participants are
randomly selected but that the variables measured constitute the population of variables. Makes it
possible to generalize from the sample to a larger population
○ Factor Rotation: you will find that most variables have high loadings on the most important factor
and small loadings on all other factors. This characteristic makes interpretation difficult, and so a
technique called factor rotation is used to discriminate factors
▪ Orthogonal Rotation (varimax, quartimax and equamax) -should never be used in Social
Sciences. Factors assumed to be independent. Values of the correlation coefficients are the
same as the values of the b-values.
▪ Oblique Rotation (direct oblimin and promax). Factors assumed to be related
○ Adequacy (KMO Bartletts Test)

Quantitative Analysis Page 3

 ○ Adequacy (KMO Bartletts Test)
○ Convergent validity (Factor loading >0.7 in Pattern Matrix)
○ Discriminant validity (No cross loadings)
○ Reliability (Cronbach's Alpha at variable/factor level)
• Confirmatory Factory Analysis
○ Obtain a roughly decent model quickly (cursory model fit, validity)
○ Do configural, metric, and scalar invariance tests (if using grouping variable in causal model)
○ Response bias (common method bias: eik factor more than 60% variance explain kar raha ho.
Check through Harman's Single Factor Test and CLF)
○ Validity (Convergent: latent to item wali loading > 0.7, Divergent: Item to Item Covariation
between 0.5 to 0.7) and Reliability check
○ Final measurement model fit
○ Optionally, impute factor scores
• Structural Equation Modeling
○ Multivariate Assumptions
▪ Outliers (Scatter Plot) and Influentials (Cook's variable, less than 1)
▪ Multicollinearity (VIF<3)
○ Hypothesis Testing should be done at:
▪ Global Level - Check overall model fit first, then go to
▪ Mid Level - Check R-Square, the go to
▪ Local Level - Standardized Regression Weight
○ Mediation
▪ Full Mediation: Only indirect effect significant
▪ Partial Mediation: Indirect + Direct path are both significant
▪ No Mediation: Indirect path is insignificant
▪ Inconsistent Mediation: path sign different than theorized
○ Moderation/Interaction
▪ Standardize constituent variables (Calculate Z-Score of all three variables)
▪ Compute new product terms (Multiply IVxMod)
▪ Plot significant interactions (Check p-value of product terms, and plot it)
○ Multigroup Comparisons
▪ Create multiple models
▪ Assign them the proper group data
▪ Test significance of moderation via chi-square difference test

Quantitative Analysis Page 4