Quantifying the New York City Subway System’s Negative Impact on Hearing

by Andrew J. Guralnick

Introduction

Millions of commuters using the New York City subway system know it can be noisy, but just how loud is it? Hearing Health Foundation (HHF) set out to measure the potential danger that the NYC Metropolitan Transit Authority (MTA) subway system presents to riders and employees.

It was found the system significantly breaches the threshold of what is safe for human ears. To protect hearing, both the U.S. Environmental Protection Agency and the World Health Organization recommend an average exposure limit of 70 decibels (dB) over the course of 24 hours. But what was measured exceeds that limit: The samples show the average noise levels on all subway platforms and on all subway rides (inside subway trains) is between 72.5 and 76.5 dB and between 74.1 and 75.8 dB, respectively. And, with maximum readings actually as high as 119 dB on platforms and 120 dB on rides—based on actual recorded data within the sample—the NYC subway is likely an auditory minefield.

Ranges were determined using the data’s sample averages to predict the actual averages are on all subway platforms and rides through the MTA system. Results are presented with a 99 percent confidence level.

Method

Data Collection

From January to August 2018, three data collectors used Decibel Meter Pro, a smartphone app, on iPhones and an iPad to collect 120 samples from platforms and rides. All 60 platform samples were equally represented at five minutes each. The 60 ride samples were assigned random recording lengths from 10 to 30 minutes.

Subway platforms were chosen mostly arbitrarily to ensure a thorough representation of the system. Sample start times were chosen randomly with a rule that the start must occur within fifty-five minutes of the designated time. For example, a 3 p.m. sample start time was permitted to begin between 2:05 and 3:55 p.m.

The start times for platforms and ride recordings were chosen utilizing a random number function in Microsoft Excel.

Random Number Generation for Platforms

Start times for recordings on platforms ranged from 5 a.m. to 10:45 p.m., Monday through Friday. The day of the week and the location on the platform was also randomly generated. The following formulas were used:

  • Start Time = RANDBETWEEN(20,91) / 4, (Military Time: .25 = :15, .5 = :30, .75 = :45)

  • Day of Week = RANDBETWEEN(2,6), (2 = Monday, 3 = Tuesday, 4 = Wednesday, 5 = Thursday, 6 = Friday)

  • Location on Platform = RANDBETWEEN(1,3), (1 = Front, 2 = Middle, 3 = Back)

Random Number Generation for Rides

Start times for recordings on train rides ranged from 5 a.m. to 10:45 p.m. and ranged from 10 to 28 minutes. The day of the week and the location on the train was also randomly generated. The starting borough and the direction were also taken into consideration.The following formulas were used:

  • Start Time = RANDBETWEEN(20,91) / 4, (Military Time: .25 = :15, .5 = :30, .75 = :45)

  • Length of Time = RANDBETWEEN(10,28).

  • Day of Week = RANDBETWEEN(2,6), (2 = Monday, 3 = Tuesday, 4 = Wednesday, 5 = Thursday, 6 = Friday)

  • Location on Ride = RANDBETWEEN(1,3), (1 = Front, 2 = Middle, 3 = Back)

  • Starting Borough = RANDBETWEEN(1,4), (1 = Manhattan, 2 = Queens, 3 = Brooklyn, 4 = Bronx)

  • Direction = RANDBETWEEN(1,4), (1 = Manhattan-bound, 2 = Queens-bound, 3 = Brooklyn-bound, 4 = Bronx-bound)

Data collectors were instructed to cease recording and leave the train at the next station after completing the assigned recording time. For example, if the data collector was assigned to record a 16-minute sample that ended between the adjacent stations of Times Square and 34th Street-Penn Station, the recording was ceased upon arrival to 34th Street, and not while the train was in motion between the two stations.

While not always successful, maximizing the assigned time to 28 minutes helped ensure that most of the samples remained under 30 minutes. A few exceptions resulted but they were included in the analysis.

The only rides under 10 minutes included are the shuttle subway lines (between Grand Central and Times Square; between Between Franklin Avenue and Prospect Park; and between Broad Channel and Rockaway Park–Beach 116th Street), as those trips are inherently shorter than 10 minutes.

Variables and Controls

The analysis examined potential harm to hearing caused by two separate concerns. The first is the concern of loud noise at subway platforms. The second is the concern of noise during subway rides. When quantified, the main independent variable for subway platforms is the number of trains which pass a subway platform, SX1. When quantified, the main independent variable for subway rides is the amount of local stations passed, TX1. The dependent variable for both cases, when quantified, was the number of seconds the subway noise level reached 75 dB or higher.

For platforms, control variables included whether a platform is above ground, whether the day is Wednesday, whether the sample is collected in the back of the platform, and whether the sample is collected during rush hour.

Surprisingly, Wednesday is predicted to be quieter than other weekdays in this analysis. A clear rationale does not exist, but the statistical significance of the relationship between this control variable and the dependent variable calls for this variable’s inclusion in the analysis. In other words, for whatever reason, Wednesday is a quieter day to use MTA subways.

For subway rides, control variables included whether a train travels between Manhattan and another borough or vice versa, whether a train runs above ground, whether the sample is collected during rush hour, and whether a local train ever becomes an express train, with fewer stops.

Statistical Methods

Ultimately, with regard to inferential statistics, two statistical methods were used. Multiple Regression was used for both subway platforms and subway rides. Also, two-sample t-tests were used to compare the averages, maximums, the percentages of time the subway noise is 75 dB or higher and the percentages of time the subway noise is 85 dB or higher.

Limitations

Among the four (of five) New York City boroughs studied—Manhattan, Queens, Brooklyn, and the Bronx—some boroughs received more coverage than others. Some weekdays received more coverage than others as well. These two limitations affect train line coverage, too. Also, three subway platform samples with missing location data were designated as “middle of platform,” as that was the most likely guess. Labeling these three platform samples as from the middle did not affect the analysis in any significant way.

Results

Multiple Regression: Platforms

Multiple regression was used for subway platforms.

Model 5 (Most Complete Model): Subway Platforms (See Table 1)

Total Time 75 dB or higher over the course of 5 minutes (In Seconds) =

140.75 + 16.53(SX1) – 90.02(SX2) – 38.87(SX3) – 18.19(SX4) – 5.80(SX5)

Each number attached to a variable represents the effect an independent variable has on the number of seconds that occurs at 75 dB or higher. “140.75” is a constant to help compute the dependent variable. The main independent variable for subway platforms—the number of trains that pass a platform, SX1—has a coefficient of 16.53 in the most complete model. This leads to the prediction that each additional train that enters or leaves a platform will add 16.53 seconds of noise at 75 dB or higher, accounting for the remaining independent variables.

For example, if a rider waits at a platform where two trains come and go before their train arrives (equal to a train passing five times), the predicted exposure to noise at 75 dB or higher is 82.65 seconds, accounting for the remaining independent control variables. Within the most complete model, the relationship between the main independent variable and the dependent variable is statistically significant at the 99 percent confidence level.

Two of four control variables, SX2  and SX3, were also statistically significant throughout the models (SX4 and SX5 were not). The first—whether a platform is above ground, SX2—has a coefficient of –90.02. This predicts that waiting at an above-ground platform for five minutes leads to 90.02 fewer seconds of 75 dB or higher, compared with waiting at a below-ground platform for five minutes. That means the five-minute wait on the above-ground platform subjects a rider to predictably 30 percent less dangerous noise. Within the most complete model, the relationship between this control variable and the dependent variable—the number of seconds the subway noise level reached 75 dB or higher—is statistically significant at a 99.9 percent confidence level.

The next control variable—whether the day of the week is Wednesday, SX3—has a coefficient of –38.87. This predicts that waiting at a platform for five minutes on a Wednesday leads to 38.87 fewer seconds of 75 dB or higher than waiting at a platform on a different weekday. That means the five-minute wait on the platform on Wednesday subjects a rider to predictably 13 percent less dangerous noise. Within the most complete model, the relationship between this control variable and the dependent variable—the number of seconds the subway noise level reached 75 dB or higher—is statistically significant at a 95 percent confidence level.

It is important to note the shift from the R2 in Model 1, 0.251, to the adjusted R2 in Model 5, 0.426. In Model 1, 25.1 percent of the variation for all seconds of noise over 75 dB can be predicted by the number of stations passed alone. In Model 5, 42.6 percent of the variation for all seconds of noise over 75 dB can be predicted by the number of stations passed as well as the control variables SX2 - SX5. This dramatic shift in R2 emphasizes the need for controls in this regression.

Multiple Regression: Subway Rides:

The statistical method of multiple regression was used for subway rides.

Model 5 (Most Complete Model): - Subway Rides (See Table 2)

Total Time 75 dB or higher (In Seconds)  =

118.49 + 36.06(TX1) + 160.22(TX2) – 54.28(TX3) + 48.90(TX4) + 67.80(TX5)

Each number attached to a variable represents an effect an independent variable has on the number of seconds that occurs at 75 dB or higher. “118.49” is a constant to help compute the dependent variable. The main independent variable for subway rides—the amount of local stations passed, TX1—has a coefficient of 36.06 in the most complete model. This leads to the prediction that each additional subway stop that is passed will add 36.06 seconds of noise of 75 dB or higher, accounting for the remaining independent control variables.

For example, if a rider passes 10 local train stops on their trip, the predicted exposure is 360.60 additional seconds—or 6.01 additional minutes—of noise at 75 dB or higher. Within the most complete model, the relationship between the main independent variable and the dependent variable—the number of seconds the subway noise level reached 75 dB or higher—is statistically significant at a 99.9 percent confidence level.

One of the four control variables, TX2—whether a train traveled between Manhattan and another borough or vice versa—was also statistically significant throughout the models (TX3, TX4, and TX5 were not). TX2 has a coefficient of 160.22. That leads to the prediction that traveling between Manhattan and another borough or vice versa leads to 160.22 additional seconds at 75 dB or higher. Within the most complete model, the relationship between this control variable and the dependent variable—the number of seconds the subway noise level reached 75 dB or higher—is statistically significant at a 95 percent confidence level.

The original R2 in Model 1 is 0.439 and the adjusted R2 in model 5 is 0.465. That means that in Model 1, 43.9 percent of the variation for all seconds of noise over 75 dB can be predicted by the number of each additional station encountered. In Model 5, 46.5 percent of the variation for all seconds of noise over 75 dB can be predicted by the number of each additional station encountered and the control variables TX2 through TX5.

Two-Sample T-Tests Assuming Unequal Variances

Two-sample t-tests were utilized in order to discover if there are statistically significant differences among specified variables between subway platforms and subway rides. The differences between platforms and subway rides with regard to dB averages, dB maximums, and percent of noise level at 75 dB or higher is not statistically significant. Nonetheless, the difference between platforms and subway rides with regard to percent of recording at 85 dB or higher is statistically significant at a 99 percent confidence level (Mean for platforms = 19.5 percent, Mean for subway rides = 3.7 percent, T-Statistic = 7.3, Two-Tailed Critical Value = 2.7, T-Statistic > Two-Tailed Critical Value). That means we are 99 percent sure that the mean for the percentage of subway platform noise at 85 dB or higher and the mean for the percentage of subway ride noise at 85 dB or higher are in fact different from each other.

The simple explanation for the discrepancy in comparisons between the two percentage-related t-tests is that many of the seconds over 75 dB recorded for subway rides fall between 75 and 84 dB, whereas many of the seconds recorded over 75 dB for platforms reach beyond 85 dB. Many more seconds recorded at 85 db or higher for subway platforms would lead to this high level of statistical significance. This result would indicate that subway platforms are potentially a lot more dangerous than subway rides.

Conclusion and Discussion

HHF’s recommendation for commuters, MTA staff, and platform retailers such as newsstand operators is simple: Wear ear protection. MTA staff and platform retailers are at elevated risk given the hours they spend underground and on the trains. The tendency for many commuters to block noise by raising the volume of their headphones is not a helpful approach and could in fact damage hearing even more.

The subway is merely only one of many sources of daily noise. Noise-induced hearing loss can result from a single, sudden noise event and from constant exposure to loud noises that has a cumulative effect (not unlike sun exposure) and can lead to related negative health effects when unknown and untreated.

The MTA appears aware of the issue of subway noise. The newly built Second Avenue subway line uses effective noise-reduction measures such as low vibration tracks and sound absorbing panels. We hope the MTA will continue to use these quieter, low vibration tracks when making subway and station upgrades, especially since they are more cost-effective than traditional wooden tracks.

2018 HHF intern Andrew J. Guralnick is pursuing a master’s in public administration at Baruch College in New York City.

Subway Platform
Variables
Coefficient Estimates (Table 1)
(Standard Errors)
Model 1 Model 2 Model 3 Model 4 Model 5
Number of Times subways Passed (Enter and Exit Separate) (SX₁) 27.09
(6.14)***
19.21
(5.80)**
17.76
(5.66)**
16.75
(5.72)**
16.53
(5.80)**
Above Ground = 1, Below Ground = 0 (SX₂) -79.06
(19.66)***
-83.15
(19.13)***
-87.47
(19.50)***
-90.02
(20.95)***
Day of Week (Wednesday = 1, Other = 0) (SX₃) -36.20
(16.53)*
-37.96
(16.58)*
-38.87
(16.91)*
Location (1 = Back, 0 = middle/front) (SX₄) -18.43
(16.82)
-18.19
(16.96)
Rush Hour, 1 = Yes, 0 = No, (6:30 - 9:30 & 15:30 - 20:00) (SX₅) -5.80
(16.49)
Constant 66.57
(22.75)**
110.75
(23.05)***
126.88
(23.50)***
137.02
(25.22)***
140.75
(27.54)***
R-sq.
(or Adjusted R²)
.251 .396 .433 .435 .426
Sample Size 60 60 60 60 60

Notes:

Significance *p<.05, **p<.01, ***p<.001

Subway Platform
Variables
Coefficient Estimates (Table 2)
(Standard Errors)
Model 1 Model 2 Model 3 Model 4 Model 5
Number of Local Stations Met (TX₁) 45.56
(6.75)***
41.39
(6.65)**
40.65
(6.81)**
39.85
(6.98)**
36.06
(5.80)**
Interborough (Between Manhattan and a different borough or vice-versa) = 1, Not = 0 (TX₂) 172.83
(66.98)*
176.03
(67.62)*
166.07
(69.99)*
160.22
(70.51)*
Above Ground = 1, Below Ground = 0 (TX₃) -44.29
(78.62)
-51.97
(80.10)
-54.28
(80.35)
Rush Hour, 1 = Yes, 0 = No, (6:30 - 9:30 & 15:30 - 20:00) (TX₄) -18.43
(16.82)
-18.19
(16.96)
Express at any point = 1, Local = 0 (TX₅) 67.80
(80.03)
Constant 143.55
(78.97)
106.95
(76.70)
123.61
(82.63)
112.51
(85.13)
118.49
(85.64)
R-sq.
(or Adjusted R²)
.439 .480 .474 .468 .465
Sample Size 60 60 60 60 60

Notes:

Significance *p<.05, **p<.01, ***p<.001