@@ -62,11 +62,38 @@ source(here::here("src", "R", "x_analysis.R"))
 
 # Abstract {.unnumbered}
 
-The ankle draft model incorporated in ASHRAE Standard 55 was derived from studies in which participants wore summer clothing with exposed ankles (~0.5 clo), potentially leading to overly conservative perimeter heating requirements in winter conditions where office occupants typically wear long trousers, socks, and closed-toe shoes (~0.75 clo). This study evaluated thermal comfort responses to ankle-level draft under such winter clothing conditions through controlled laboratory experiments with 51 participants. Using a within-subjects repeated-measures design, we exposed participants to nine conditions varying supply air temperature (15°C, 17°C, 19°C) and ankle-level air speed (0.1–0.7 m/s) simulating window downdraft while they wore clothing ensembles targeting 0.75 clo.
-
-The existing ankle draft model systematically overpredicted dissatisfaction across all conditions. While prior studies with exposed ankles reported dissatisfaction rates of 23–57%, we observed rates of only 6–20%, with the existing model overestimating dissatisfaction by an average of 2.8 times and by as much as 39 percentage points at high air speeds. We fit generalized linear mixed-effects models to data from this study and prior research, developing separate prediction equations for ankle-covered and ankle-uncovered conditions. The covered-ankle model exhibited both a lower baseline dissatisfaction risk and reduced sensitivity to air speed. Whole-body thermal sensation and ankle-level air speed were the only significant predictors; ankle air temperature and room temperature were not significant within the tested ranges.
-
-These findings support differentiating ASHRAE Standard 55 ankle draft criteria by clothing insulation level. Adopting a covered-ankle criterion for winter heating scenarios will enable meaningful reductions in perimeter heating requirements without compromising occupant comfort, with associated benefits for installation costs, energy consumption, and carbon emissions.
+The ankle draft model incorporated in ASHRAE Standard 55 was derived
+from studies in which participants wore summer clothing with exposed
+ankles (\~0.5 clo), potentially leading to overly conservative perimeter
+heating requirements in winter conditions where office occupants
+typically wear long trousers, socks, and closed-toe shoes (\~0.75 clo).
+This study evaluated thermal comfort responses to ankle-level draft
+under such winter clothing conditions through controlled laboratory
+experiments with 51 participants. Using a within-subjects
+repeated-measures design, we exposed participants to nine conditions
+varying supply air temperature (15°C, 17°C, 19°C) and ankle-level air
+speed (0.1–0.7 m/s) simulating window downdraft while they wore clothing
+ensembles targeting 0.75 clo.
+
+The existing ankle draft model systematically overpredicted
+dissatisfaction across all conditions. While prior studies with exposed
+ankles reported dissatisfaction rates of 23–57%, we observed rates of
+only 6–20%, with the existing model overestimating dissatisfaction by an
+average of 2.8 times and by as much as 39 percentage points at high air
+speeds. We fit generalized linear mixed-effects models to data from this
+study and prior research, developing separate prediction equations for
+ankle-covered and ankle-uncovered conditions. The covered-ankle model
+exhibited both a lower baseline dissatisfaction risk and reduced
+sensitivity to air speed. Whole-body thermal sensation and ankle-level
+air speed were the only significant predictors; ankle air temperature
+and room temperature were not significant within the tested ranges.
+
+These findings support differentiating ASHRAE Standard 55 ankle draft
+criteria by clothing insulation level. Adopting a covered-ankle
+criterion for winter heating scenarios will enable meaningful reductions
+in perimeter heating requirements without compromising occupant comfort,
+with associated benefits for installation costs, energy consumption, and
+carbon emissions.
 
 # Introduction
 
@@ -164,7 +191,13 @@ whole-body thermal sensation of approximately −0.5 (slightly cool) on
 the ASHRAE seven-point scale as calculated using the CBE Comfort Tool
 [@tartariniCBEThermalComfort2020].
 
-![Experimental configuration and temporal protocol: (a) Seated occupant positioned 0.7 m from the custom displacement diffuser, which simulates window downdraft by directing conditioned air toward the ankle region from behind; (b) Timeline of the 120-minute experimental sessions featuring overlapping participant groups and synchronized questionnaire intervals to characterize the progression of local and whole-body thermal perception.](figs/experimental_setup.png){#fig-setup
+![Experimental configuration and temporal protocol: (a) Seated occupant
+positioned 0.7 m from the custom displacement diffuser, which simulates
+window downdraft by directing conditioned air toward the ankle region
+from behind; (b) Timeline of the 120-minute experimental sessions
+featuring overlapping participant groups and synchronized questionnaire
+intervals to characterize the progression of local and whole-body
+thermal perception.](figs/experimental_setup.png){#fig-setup
 width="75%"}
 
 We used custom displacement diffusers to deliver conditioned air to the
@@ -305,7 +338,7 @@ calibration intercept and slope, AUC, and Brier score. All analyses were
 implemented using the R package CalibrationCurves
 [@deCockCalibrationCurves2023].
 
-### Development of the new ankle draft model
+### Development of the new ankle draft model {#sec-method-new-model}
 
 We modeled ankle draft dissatisfaction using a generalized linear mixed
 model with a binomial distribution and logit link, as the outcome was
@@ -732,7 +765,7 @@ intensified as air speed increased, with the most significant
 temperature drops occurring when moving from low to medium
 airflow.](figs/tsk_p.png){#fig-tsk width="90%"}
 
-## Dissatisfaction with ankle draft
+## Dissatisfaction with ankle draft {#sec-model-comp}
 
 ```{r}
 #| label: setup-results-data
@@ -779,24 +812,11 @@ pp.
 (a) The model consistently overpredicted ankle-level dissatisfaction
 across all conditions, with discrepancies widening at higher air speeds
 (points represent mean values, with the segment length indicating the
-magnitude of model overestimation); (b) while maintaining acceptable
-discrimination (AUC =
-`r sprintf("%.2f", liu_calibration_stats["C (ROC)"])`), the calibration
-curve confirms a systematic overestimation of dissatisfaction in winter
-conditions.](figs/model_error_combined.png){#fig-model-comp width="90%"}
-
-In addition, as shown in @fig-model-comp(b), the model retained
-acceptable discrimination on this dataset (AUC =
-`r sprintf("%.3f", liu_calibration_stats["C (ROC)"])`), the calibration
-results indicated a systematic overestimation under winter conditions.
-Across the nine experimental conditions, the mean observed
-dissatisfaction rate was `r liu_model_summary$mean_observed`%, whereas
-the Liu model predicted an average of
-`r liu_model_summary$mean_predicted`%, corresponding to an
-overestimation of approximately
-`r liu_model_summary$overprediction_ratio` times. The discrepancy was
-greatest at `r liu_max_diff_speed` air speed and smaller, but still
-evident, at `r liu_min_diff_speed` air speed.
+magnitude of model overestimation); (b) calibration curve of the existing model on this study dataset.](figs/model_error_combined.png){#fig-model-comp width="90%"}
+
+In addition, we evaluated the performance of the existing ankle draft model by fitting its predictions—based on local air speed and whole-body thermal sensation—against our observed binary satisfaction outcomes using logistic regression. The resulting calibration curve is shown in @fig-model-comp(b). In a well-calibrated model, the curve should closely follow the 45° ideal line; however, our results show the curve located predominantly below this line.
+
+While the model maintained acceptable discriminative ability (AUC = 0.77), meaning it can effectively rank which conditions are more likely to cause discomfort, the calibration statistics reveal significant inaccuracy in the magnitude of its predictions. With a Brier score of 0.14 and a systematic shift away from the ideal line, the results confirm that the existing model consistently overestimates the probability of ankle draft dissatisfaction under winter conditions.
 
 ## Updated model for winter conditions
 
@@ -902,15 +922,22 @@ model_final_table <- dplyr::bind_rows(
   dplyr::select(model_display, term_label, estimate_fmt, std_error_fmt, z_value_fmt, p_value_fmt)
 ```
 
-As shown in @tbl-updated-model-screening, in the mixed-effects model
-including all candidate predictors, air speed at ankle, whole-body
-thermal sensation, and ankle condition were significantly associated
-with draft dissatisfaction at the ankle, whereas room air temperature,
-air temperature at ankle, and turbulence intensity were not significant
-within the range investigated in this study. Based on these results,
-separate mixed-effects models for ankle draft dissatisfaction were
-developed for the ankle-uncovered and ankle-covered conditions, using
-air speed at ankle and whole-body thermal sensation as predictors.
+As shown in @sec-model-comp, the existing model showed limited accuracy
+in predicting ankle draft dissatisfaction under winter conditions. This
+highlights the need for a revised model that accounts for differences in
+application contexts. To enable an integrated analysis of the effects of
+different factors on ankle draft dissatisfaction, we combined data from
+@liuPredictedPercentageDissatisfied2017 and collected in this study
+in the modeling dataset. To identify significant predictors, a model
+including all candidate predictors was first fitted following the
+methods described in @sec-method-new-model. Given that both datasets
+were obtained from repeated measurements, subject-level dependence was
+accounted for in the analysis. The result of the screening model is
+shown in @tbl-updated-model-screening. Ankle-level air speed, whole-body
+thermal sensation, and ankle condition were significantly correlated
+with ankle draft dissatisfaction, whereas room air temperature,
+ankle-level air temperature, and turbulence intensity were not
+significant within the range investigated in this study.
 
 ```{r}
 #| label: tbl-updated-model-screening
@@ -934,15 +961,17 @@ knitr::kable(
 )
 ```
 
-As shown in @tbl-updated-model-final, the two final models exhibited the
-same structure but different coefficients. The intercept of the covered
-model was `r sprintf("%.2f", abs(glmm_delta_intercept))` units lower
-than that of the uncovered model, indicating a substantially lower
-baseline risk of ankle draft dissatisfaction when the ankle was covered
-by clothing. In addition, the air speed coefficient was smaller in the
-covered model than in the uncovered model, indicating reduced
-sensitivity of dissatisfaction to air velocity under ankle-covered
-conditions.
+Based on these results, separate mixed-effects models for ankle draft
+dissatisfaction were developed for the ankle-uncovered and ankle-covered
+conditions, using ankle-level air speed and whole-body thermal sensation
+as predictors. As shown in @tbl-updated-model-final, the two final
+models exhibited the same structure but different coefficients. The
+intercept of the covered model was 1.91 units lower than that of the
+uncovered model, indicating a lower baseline risk of ankle draft
+dissatisfaction when the ankle was covered by clothing. In addition, the
+air speed coefficient was smaller in the covered model than in the
+uncovered model, indicating reduced sensitivity of dissatisfaction to
+air speed under ankle-covered conditions.
 
 For the ankle-uncovered condition, the approximate closed-form equation
 derived from the marginal predictions of the mixed-effects model can be