docs: clarify demonstrative pronouns and improve narrative flow

_subfiles/Linear-models-overview/_exr-linreg-rescaling.qmd

@@ -127,7 +127,7 @@ Slope coefficients ($\g_{A^*}$ and $\g_{A^*M}$) are unchanged by rescaling:
 they have the same values and interpretations as the corresponding $\b$s.
 
 Coefficients change only for variables that have interactions with the rescaled variable $A$.
-This includes the intercept
+These affected parameters include the intercept
 (which can be viewed as the main effect of a variable that interacts with $A$ via $\b_A$),
 and the main effect of $M$
 (which interacts with $A$ via $\b_{AM}$).

_subfiles/Linear-models-overview/_sec_hers_lm_diagnostics_lm2.qmd

@@ -50,7 +50,7 @@ The residual and QQ plots for the interaction model
 (@fig-hers-resid-fitted-lm2, @fig-hers-qq-lm2)
 look essentially identical to those for the parallel-planes model:
 adding the age-BMI interaction barely changes the residual structure.
-This previews the model-selection result ---
+This similarity previews the model-selection result ---
 the likelihood ratio test (@tbl-hers-lrt) shows the interaction
 does not improve the fit enough to justify the extra parameter.
 :::

_subfiles/Linear-models-overview/_sec_linreg_model_selection.qmd

@@ -28,7 +28,7 @@ rather than only with numerical fit.
 In this workflow,
 the DAG encodes hypothesized time ordering
 and causal pathways.
-That helps us decide
+That structure helps us decide
 which variables belong in the candidate model set
 before we run stepwise,
 subset,

_subfiles/proportional-hazards-models/_sec-competing-risks.qmd

@@ -254,7 +254,7 @@ event probability because many men are still event-free at 5 years.
 A naive approach treats competing events as censored
 and uses the standard Kaplan-Meier method:
 $1 - \hskmf{t}$.
-This overestimates the true cumulative incidence
+This naive Kaplan-Meier approach overestimates the true cumulative incidence
 because it assumes censored individuals
 (including those who died)
 have the same risk as those still event-free.