topic 4

Overview

  • Topic 4 covers essential concepts in econometrics, focusing on different ways to model relationships between economic variables (functional forms), how we measure those variables (units of measurement), and what these choices mean for understanding data and its real-world implications. These concepts are crucial for accurately interpreting regression results and making informed economic decisions.

Chapter References

  • Relevant chapters to refer to include:

    • Chapter 6.2: Slope coefficients interpretation

    • Chapter 6.4: Semi-log (or log-lin) model: semi-elasticities

    • Chapter 6.5: Lin-log model: semi-elasticities

    • Chapter 6.6: Polynomial equations

    • Chapter 4.1: Scaling and units of measurement

Slope Coefficients Interpretation (Section 4.1)

  • The marginal effect of one variable (X) on another (Y) is simply the slope of the relationship between them. This slope tells us how much Y changes for a small change in X. Mathematically, it's calculated as a derivative, \frac{dY}{dX}, which measures the instantaneous rate of change.

  • It is crucial to observe how variables are entered into the equation: either as their original values (natural units) or transformed using a natural logarithm (ln).

    • Linear Model Definition: This is the simplest model where variables are in their natural units.

      Yi = \beta1 + \beta2 Xi + u_i

      • Here:

        • Y_i represents the dependent variable (the outcome we are trying to explain) for observation i.

        • X_i represents the independent variable (the factor we believe influences Y) for observation i.

        • \beta_1 is the intercept, indicating the value of Y when X is zero.

        • \beta_2 is the slope coefficient, which is the primary focus of interpretation.

        • u_i represents the error term, accounting for unobserved factors or random noise.

    • Interpretation of \beta2: In this linear model, \beta2 tells us the actual number of units that Y is expected to change for every one-unit increase in X, assuming all other factors remain constant.

Semi-log (Log-Lin) Model (Section 4.2)

  • Log-Lin Model Definition: In this model, both the dependent variable (Y) and the independent variable (X) are transformed using their natural logarithms.

    \ln Yi = \beta1 + \beta2 \ln Xi + u_i

    - The natural logarithm (often written as `ln`) helps us interpret changes in terms of percentages rather than absolute units.

    • Interpretation of \beta2: Here, \beta2 is called the elasticity of Y with respect to X. This means it represents the approximate percentage change in Y that results from a one percent change in X. For example, if \beta_2 is 0.5, a 1% increase in X leads to a 0.5% increase in Y.

  • The transformation from absolute changes to relative (percentage) changes can be understood as follows:

    • When we talk about changes in variables:

      • Change in Y: dY refers to a change in Y expressed in its original units.

      • Change in X: dX refers to a change in X expressed in its original units.

      • Change in \ln(Y): d\ln(Y) is approximately equal to \frac{dY}{Y} which represents the relative change (or percentage change if multiplied by 100) in Y.

      • Change in \ln(X): d\ln(X) is approximately equal to \frac{dX}{X} which represents the relative change (or percentage change if multiplied by 100) in X.

    • Therefore, when both sides of the equation are in logarithms, the coefficient \beta_2 directly captures the ratio of the relative change in Y to the relative change in X.

Interpretation of Semi-elasticity in Semi-log Models

  • In the semi-log model, only the dependent variable (Y) is transformed using its natural logarithm, while the independent variable (X) remains in its natural units.

    • Model Definition:

      \ln(Yt) = \beta1 + \beta2 Xt + u_t

    • Interpretation for \beta2: In this case, \beta2 represents the approximate relative change in Y (as a decimal) for a one-unit increase in X. To express this as a percentage, you multiply \beta_2 by 100.

      • So, \beta2 \times 100 gives the percentage change in Y due to a one-unit increase in X. For example, if \beta2 is 0.04, a one-unit increase in X leads to an approximately 4% increase in Y.

Practical Example of Semi-log Model

  • Consider a model used to analyze the growth of an economy's Gross Domestic Product (GDP) over time:

    • Model Used for GDP: \ln(GDPt) = \beta1 + \beta2 t + ut

    • Here, t denotes time, typically measured in years.

    • Interpretation of \beta2: In this context, \beta2 is directly interpreted as a growth rate. For example, if \beta_2 is 0.04:

      • An increase of one year (t increases by 1) results in an approximate (0.04 \times 100)\% = 4\% growth rate in GDP. This means GDP is expected to increase by 4% each year.

Lin-log Model (Section 4.3)

  • This model is the reverse of the semi-log model: the dependent variable (Y) is in natural units, but the independent variable (X) is transformed using its natural logarithm.

    • Lin-Log Model Definition:

      Yi = \beta1 + \beta2 \ln(Xi) + u_i

    • Interpretation of \beta2: Here, \beta2 measures the units change in Y resulting from a one percent increase in X. Specifically, for a 1% increase in X, Y is expected to change by \frac{\beta_2}{100} units.

  • Example: Let's look at expenditure on services as a function of total personal consumption expenditure:

    • \text{Expenditure on services}t = \beta1 + \beta2 \ln(\text{Total personal consumption}t) + u_t

    • Empirical Findings from 1975-2006: If a study found that \beta2 was 18.44, then a 1% increase in total personal consumption is associated with an increase of \frac{18.44}{100} = 0.1844 units in expenditure on services. If Y is measured in billions of dollars, this would mean an increase of $0.1844 billion, or $184.4 million, in expenditure on services, not $18.44 billion as implied in the original note's example. To clarify for the note, if \beta2 is 1844 (for example) then a 1% increase in consumption increases expenditure by $18.44 billion.

Polynomial Equations (Section 4.4)

  • Polynomial functions are used when the relationship between variables is not a simple straight line but rather curves. They are frequently employed in economic contexts such as modeling cost functions (how total cost changes with production) or production functions (how output changes with input).

  • Quadratic Function Definition: This involves X and X squared (X^2), creating a single bend or curve.

    Yi = \beta1 + \beta2 Xi + \beta3 Xi^2 + u_i

  • Cubic Function Definition: This includes X, X squared (X^2), and X cubed (X^3), allowing for two bends in the curve.

    Yi = \beta1 + \beta2 Xi + \beta3 Xi^2 + \beta4 Xi^3 + u_i

  • Insights on the effect of X on Y when using quadratic terms (only consider \beta_3 for the quadratic relationship):

    • If \beta_3 > 0: The model exhibits a U-shape (convex). This means Y initially decreases as X increases, reaches a minimum point, and then starts to increase. A common example is an average cost curve, which initially falls due to economies of scale and then rises due to diminishing returns.

    • If \beta_3 < 0: The model exhibits an inverted U-shape (concave). This means Y initially increases as X increases, reaches a maximum point, and then starts to decrease. An example is an average product curve, where adding more input initially boosts productivity but eventually leads to reduced average output.

  • The stationary point (maximum or minimum of Y) is determined by the first-order condition (F.O.C). This is found by setting the first derivative of the function with respect to X to zero:

    • \frac{\delta Y}{\delta X} = 0

    • In simpler terms, we are finding the point on the curve where the slope is perfectly flat (zero). This flat point indicates either the highest (maximum) or lowest (minimum) value of Y along that curve, given the values of X.

Implications of Scaling and Units of Measurement (Section 4.5)

  • Regression analysis is flexible and allows data to be expressed in various measurement units, such as individual units, thousands, millions, or billions. While convenient, it has important implications for interpreting regression results.

  • Important implications regarding Regression results:

    • The estimated coefficients (intercepts and slopes) will change if the units of measurement for X or Y are changed. However, the fundamental interpretation of the underlying economic relationship, the R-squared (which measures how well the model fits the data), and the statistical significance of the coefficients remain the same.

    • Example: Analyzing pints of beer sold (Y) based on price (X).

      • Suppose a model gives us the following coefficients:

        • \beta_2 = -85: This means if the price (X) increases by $1 (assuming X is in dollars), the consumption of beer (Y) is expected to decrease by 85 pints.

        • \beta_1 = 1200: This is the consumption in pints when the price is zero (though this might not be economically relevant, it's the intercept).

      • Conversion of price to pence: If we re-estimate the model but measure price in pence instead of dollars ($1 = 100$ pence), the new slope coefficient (\gamma_2) would adjust.

        • The new model might look like: Yi = \gamma1 + \gamma2 X^pi + u_i (where X^p is price in pence).

        • The new slope \gamma2 would be calculated as \frac{\beta2}{100}. So, \gamma_2 = \frac{-85}{100} = -0.85. This means if the price increases by 1 penny, consumption would decline by 0.85 pints. The impact is the same; it's just scaled correctly for the new unit.

      • Changing Y from pints to half-pints: If we decide to measure consumption in half-pints (Y^2i) instead of pints (Yi), where Y^2i = 2 \times Yi, the coefficients would again adjust.

        • The new model might be: Y^2i = \delta1 + \delta2 Xi + ui. Here, \delta1 = 2 \times \beta1 and \delta2 = 2 \times \beta_2. So, an increase of $1 in price would now lead to a decrease of 2 \times 85 = 170 half-pints.

  • These examples demonstrate how changes in the units of measurement for either the dependent or independent variables lead to corresponding adjustments in the regression coefficients. While the numerical values of the coefficients change, the underlying economic relationship and the interpretation of the impact (e.g., that an increase in price reduces consumption) remain consistent. This understanding is crucial for correctly interpreting models, especially when dealing with large numbers like government expenditures over time, where units (pounds, millions, billions) can drastically alter coefficient values but not the fundamental economic insight like growth metrics or the trend in governmental expenditure analysis.