91Ó°ÊÓ

A quantitative variable is one that represents numeric values, making it possible to perform arithmetic operations on them. In our example, the number of older siblings is a discrete quantity ranging from 0 to 4. Each number has a unique meaning, reflecting an actual count of siblings. Including this variable as a quantitative variable in a regression model assumes a direct linear relationship between the number of older siblings and the maturation of eighth graders. This approach is straightforward as it treats the variable as a continuous scale. For example, going from 0 to 1 sibling or from 3 to 4 siblings is assumed to have an equivalent impact on maturation. This simplicity, however, may not capture more complex, non-linear interactions hidden within the data.

Indicator variables, also known as dummy variables, are used to represent categorical data in regression models. These variables can take on a value of either 0 or 1. For the number of older siblings variable, we can create four indicator variables corresponding to the values from 1 to 4 (e.g., 'Has_1_Sibling', 'Has_2_Siblings', etc.). For any given observation, only one of these indicators will be 1, and the others will be 0. This approach allows the model to treat each sibling count as a distinct category rather than maintaining a linear assumption. This is useful if, for example, the maturation effect is significantly different between having no siblings and having one sibling, or between having two siblings and having three siblings. While this method can capture non-linear effects and categorical differences, it increases the model’s complexity because it adds more parameters.

A linear relationship means that there is a straight-line association between the predictor variable and the outcome variable. When we treat the number of older siblings as a quantitative variable, we assume that each additional sibling adds a constant amount to the maturation level of the eighth graders. Mathematically, this can be expressed as: \[ y = \beta_0 + \beta_1 x + \text{error} \] where \( y \) is the maturation level, \( x \) is the number of older siblings, and \( \beta_1 \) is the coefficient representing the change in \( y \) for each one-unit change in \( x \). If the true relationship is not linear, then this model may not fit the data well, leading to poor predictions and interpretations. It's crucial to explore whether linearity is a reasonable assumption by plotting the data and checking for deviations.

Non-linear effects occur when changes in the predictor variable lead to non-constant changes in the outcome variable. For instance, having one older sibling might have a different impact on maturation compared to having three or four older siblings. When we use indicator variables, we allow for the possibility of non-linear relationships in our regression model. In this scenario, the model does not assume that the impact of going from 0 to 1 sibling is the same as going from 3 to 4 siblings. Instead, each scenario (e.g., 'Has_1_Sibling', 'Has_2_Siblings') can affect the maturation level differently. This flexibility can better capture the reality of how sibling numbers influence maturation if the relationship is more complex than a simple line. Remember, realizing non-linear effects potentially improves the model's accuracy but makes it more complex to interpret.