Accuracy and Consistency of Confidence Limits for Monosyllable Identification Scores Derived Using Simulation, the...

Area of Science:

• Audiology
• Speech Perception
• Statistical Modeling

Background:

• Assessing maximum speech identification scores (PBmax) relative to pure-tone average (PTA) thresholds is crucial for audiological diagnosis and rehabilitation.
• Accurate estimation of the 95% confidence limit (CL) for PBmax is needed to determine if scores are poorer than expected based on PTA.
• This study evaluates three methods for estimating the 95% CL of PBmax scores.

Purpose of the Study:

• To compare the accuracy and consistency of three methods for estimating the 95% confidence limit (CL) of maximum speech identification scores (PBmax).
• To identify the most reliable method for clinical audiological practice.

Main Methods:

• Three methods were compared: a simulation method, the Harrell-Davis (HD) estimator, and non-linear quantile regression (nQR).
• The HD and nQR methods are distribution-free and do not require pre-defined pure-tone average (PTA) subgroups.
• Accuracy and consistency were assessed by applying each method to random data samples and predicting remaining data.

Main Results:

• The simulation method exceeded the target 5% false positive rate, with 6.7-8.2% of scores falling below the estimated 95% CL.
• The Harrell-Davis (HD) and non-linear quantile regression (nQR) methods demonstrated good accuracy, with approximately 5% of scores falling below the 95% CL.
• All methods showed similar consistency in their estimations.

Conclusions:

• The non-linear quantile regression (nQR) method is recommended for estimating the 95% confidence limit of maximum speech identification scores.
• nQR offers superior accuracy and consistency compared to the simulation method and does not require arbitrary pure-tone average subgroups.
• This method can improve audiological diagnosis and rehabilitation by providing a more reliable benchmark for speech identification performance.