RTI Press
Choosing the Best Mattress:
An Experiment in Testing
Whether Individuals
Choose a Bed at Leads to
Improved Sleep
Sean O. Hogan, Jack D. Edinger, Gayle S. Bieler, and
Andrew D. Krystal
August 2011
RESEARCH REPORT
This publication is part of the
RTI Press Research Report series.
RTI International
3040 Cornwallis Road
PO Box 12194
Research Triangle Park, NC
27709-2194 USA
Tel: +1.919.541.6000
Fax: +1.919.541.5985
E-mail: rtipress@rti.org
Web site: www.rti.org
RTI Press publication RR-0016-1108
is PDF document was made available from www.rti.org as a public service
of RTI International. More information about RTI Press can be found at
http://www.rti.org/rtipress.
RTI International is an independent, nonprot research organization dedicated
to improving the human condition by turning knowledge into practice. e
RTI Press mission is to disseminate information about RTI research, analytic
tools, and technical expertise to a national and international audience. RTI Press
publications are peer-reviewed by at least two independent substantive experts
and one or more Press editors.
Suggested Citation
Hogan, S. O., Edinger, J. D., Bieler, G. S., and Krystal, A., D. (2011). Choosing the
best mattress: An experiment in testing whether individuals choose a bed that leads
to improved sleep (RTI Press publication No. RR-0016-1108). Research Triangle
Park, NC: RTI Press. Retrieved [date] from http://www.rti.org/rtipress.
©2011 Research Triangle Institute. RTI International is a trade name of Research Triangle
Institute.
All rights reserved. Please note that this document is copyrighted and credit must be provided to
the authors and source of the document when you quote from it. You must not sell the document
or make a prot from reproducing it.
doi:10.3768/rtipress.2011.rr.0016.1108
www.rti.org/rtipress
About the Authors
Sean O. Hogan, PhD, is a former
project director in RTI International’s
Survey Research Department. He
participated in the design of the
study and served as its project
director between 2008 and 2011.
Jack D. Edinger, PhD, is a senior
psychologist at the Veterans Affairs
Medical Center in Durham, North
Carolina.
Gayle S. Bieler, MS, is a senior
statistician at RTI International. Ms.
Bieler led design and implementation
of statistical data analysis.
Andrew D. Krystal, MD, is director
of the Quantitative EEG Laboratory
at Duke University Medical Center,
and an assistant professor in Duke’s
Department of Psychiatry and
Behavioral Sciences.
Choosing the Best Mattress:
An Experiment in Testing Whether
Individuals Choose a Bed at Leads to
Improved Sleep
Sean O. Hogan, Jack D. Edinger, Gayle S. Bieler, and
Andrew D. Krystal
Abstrac t
A comfortable mattress is said to be an essential ingredient in a good night’s
sleep, but we have little understanding of the effects of sleep surface on sleep
outcomes such as daytime drowsiness or energy. Most studies devoted to testing
the effects of sleep surface on sleep have been hampered by methodological
shortcomings; these include having small numbers of subjects and evaluating
a narrow array of bedding systems. We hypothesized that motion and self-
reported measures of sleep quality and outcomes would demonstrate that the
optimal mattress would differ from person to person. We hypothesized that
individuals would be able to select one mattress from among several under
showroom circumstances that would lead to optimal rest. We find that optimal
mattress firmness varies among individuals and is reflected, at least to a degree,
by overnight motion. When allowed to test mattresses in a typical showroom
experience, individuals choose a mattress that does not minimize overnight
motion and maximize perceived sleep quality. This suggests that they may not
be receiving the health benefits that come from optimal rest. Therefore, both
manufacturers and sleep scientists could improve sleep outcomes by testing
ways to help consumers select a mattress.
Contents
Introduction 2
Methods 3
Study Design and Sample 3
Statistical Methods 3
Results 6
Actigraphically Determined
Bed Ranking (Motion Bed
Rank) 6
Sleep Quality Bed Ranking 8
Self-Selected Versus
Actigraphically Determined
Best Bed 9
Discussion 10
References 11
Acknowledgments Inside back cover
2 Hogan et al., 2011 RTI Press
Introduction
A growing body of literature indicates the eects
of sleep on health, ability to function, and quality
of life (Alapin et al., 2001; CDC, 2007; Elmenhorst
et al., 2008; Hamilton et al., 2007; Hanel, Dartman,
& Shishoo, 1996; NIH, 2003a, 2003b; Roberts,
Roberts, & Duong, 2008; Van Dongen, Maislin,
Mullington, & Dinges, 2003). A comfortable mattress
is commonly assumed to be an essential ingredient
in a good night’s sleep (Better Sleep Council,
2008b). However, we have little understanding of
the eects of sleep surface on sleep outcomes such
as daytime drowsiness or energy. Moreover, most
studies devoted to testing the eects of sleep surface
on sleep have been hampered by methodological
shortcomings; these include having small numbers
of subjects and evaluating a narrow array of bedding
systems (Lopez-Torrez, Porcar, Solaz, & Romero,
2008).
When trying to select a comfortable mattress, or
one that will provide optimal sleep, consumers
confront the decision to choose from among several
mattresses. Sources of information easily accessible
to laypersons (see for example reports by the Better
Sleep Council, 2008b; and Consumer Reports,
2005) essentially tell consumers to trust their own
judgment. In other words, consumers are told to base
their decision on an in-store experience of lying on,
sitting on, and feeling the mattresses. Unfortunately,
the scientic literature is not much more helpful
to health care providers who might wish to oer
guidance to the layperson choosing a mattress. is
literature has been silent on the extent to which
individuals should evaluate a mattress, or whether
they are capable of selecting the mattress that leads
to best outcomes, such as quality of rest, reduced
drowsiness, or increased daytime energy. In addition,
we found that the current literature typically suers
from three shortcomings that have deprived experts
and laypersons alike of this knowledge (Krystal,
Edinger, Bieler, Mladsi, & Hogan, 2011). ese
deciencies are that most studies of mattress eects
on sleep have relied on (1) a small number of people
enrolled in the study, (2) a narrow array of test
mattresses,
1
and (3) a narrow focus on individuals
suering from a chronic sleep ailment of some sort.
To help address this deciency in knowledge, we
recruited a sample of 128 healthy adults (referred to
as subjects, participants, or individuals) and asked
them to sleep on an array of seven mattresses for up
to 1 month each. We recorded measures of overnight
motion (with a sensor called an actigraph), and
participants completed diary reports of sleep quality
and daytime function. According to the collected
data, individuals vary substantially in the degree of
mattress rmness that reduces their morning pain
and optimizes their sleep experience and subsequent
daytime functioning (Krystal et al., 2011). Along
with validating the methods of measurement, we
reported that a slight increase in sleep eciency
(actigraphically measured time devoted to sleep
actually spent sleeping) can lead to improved quality
of sleep as reported in diary observations.
In this, our second report from this study, we build
on the foundation of our earlier paper (Krystal et al.,
2011). For this study, we hypothesized that overnight
actigraphic motion measurements (which indicate
sleep and awake periods during the time devoted
to sleep) and self-reported diary measures of sleep
would demonstrate that an optimal mattress could
be identied for an individual and that the “best”
mattress would dier from person to person. We also
hypothesized that changes in measured overnight
motion would coincide with other measures of sleep
and daytime functioning (e.g., as daytime energy and
drowsiness). Finally, we hypothesized that individuals
would be able to rely on conventional shopping
procedures to select one mattress from among several
under showroom circumstances that would lead to
optimal rest. Here we address these questions. As
in the last paper, we relied on a randomized, single-
blind, within-subject crossover study examining
multiple levels of mattress rmness in a large sample
of individuals without complaints of pain or sleep
diculty.
1
In this report, the terms mattress and bed are used interchangeably.
Choosing the Best Mattress 3
Methods
Study Design and Sample
Our previous paper explains in detail our study
procedures, sample selection, and methods of
measurement. We summarize our methods here to
inform readers of the basic elements of our protocol.
We recruited a convenience sample of 128 healthy
adults who lived in the Raleigh-Durham area of
North Carolina. Table 1 describes the age, body mass
index (BMI), gender, race, and partnership status
of our sample members. None of the subjects had a
sleep-aecting disease, sleep-disrupting prescription
drug regimen, or lifestyle that is known to interrupt
sleep (e.g., frequent travel, infants to care for,
overnight work shi).
(Ancoli-Israel et al., 2003; Morgenthaler et al., 2007).
e American Academy of Sleep Medicine supports
the use of actigraphic measurement to identify sleep
and wake periods (Littner et al., 2003). To measure
pain, sleep quality, daytime drowsiness, and other
parameters, participants entered reports in an
electronic diary.
Self-Selection of Preferred Mattress
Before beginning the in-home part of the sleep
study, we sought to test whether the typical shopping
experience would lead subjects to choose the one
mattress that results in best sleep. We simulated a
showroom environment where all of the subjects tried
each mattress. To maintain the blind study protocol,
we arranged the seven study mattresses in random
order in the simulated showroom. e participants
were blinded to the mattress’s manufacturer,
construction materials, design, and level of rmness.
We asked them to act as though they were in the
market to buy a mattress and to select the one
they preferred. As part of the selection process,
we encouraged each participant to “test drive” the
mattress. ey were encouraged to lie on, feel, and
evaluate the mattresses. Participants were allowed
as much time as they wanted to make their selection
(typically they took 10 to 15 minutes) and were able
to make notes of their observations. We conducted
the mattress self-selection during participant training
in a laboratory located in Research Triangle Park,
North Carolina.
Statistical Methods
We began the analysis by ranking the mattresses
(referred to in the rest of this section as “beds”), at the
participant level, according to their average amount
of overnight motion on each bed. is was measured
in terms of the number of minutes during which
the actigraph measured motion during the night,
normalized to 8 hours devoted to sleep. e measure
of motion refers to the number of 1-minute intervals
in which the actigraph measured any amount of
motion, during the time from sleep onset to nal
arousal, while the subject was lying in bed. We ranked
the beds for each individual from best (coded 1)
to worst (coded 7), where 1 indicates that Bed j
(j=1,…,7) has least average actigraphic motion for
Table 1. Descriptive statistics of study sample
Variable Sample Size Mean or Percentage Distribution
Age in years 128 40.4 (Range: 24.0–68.0)
BMI 128 25.9 (Range: 17.9–45.0)
Gender 128 61% Female
39% Male
Race/ethnicity 128 80.5% White
13.3% African American
5.5% Asian
0.8% Native American
Partner status 128 66% slept with partner
41% were members of couples in
the study (26 couples in study)
BMI = body mass index.
Subjects slept on each of the seven test mattresses
in their own homes. e mattresses were made by
the same manufacturer, ensuring consistency in
materials and production processes so that the only
dierence evaluated would be rmness. e inner-
spring mattresses ranged in rmness to mirror the
range typically found in the US marketplace. Each
of the subjects in this study used each mattress for
approximately 1 month. ey were assigned to each
mattress using a Latin square randomization system
so that subjects were not on consecutively rmer or
soer mattresses. Nothing on the mattress would have
indicated to the subject the level of rmness when the
bed arrived at the home.
To measure sleep duration and eciency, participants
wore a widely accepted monitor called an actigraph
4 Hogan et al., 2011 RTI Press
subject i, and 7 indicates Bed j has greatest actigraphic
motion for subject i. We call the eect resulting from
this mattress ranking the motion bed rank. Based on
this ranking, a participant’s best motion bed is the
mattress with the smallest average motion minutes
per night (normalized to 8 hours devoted to sleep).
We used a linear regression model to estimate and
compare average overnight motion within each
motion bed rank category. is analysis estimates
the degree to which motion (reported in minutes)
was reduced on the best motion bed as compared
with other beds. To conduct the linear regression
analyses, we analyzed longitudinal data from
all compliant nights on all mattresses from each
participant simultaneously in a general linear mixed
model (Diggle, Liang, & Zeger, 1994; Laird & Ware,
1982; Lindsay, 1993). e outcome variable was
overnight motion. e main independent variable
of interest—motion bed rank—was modeled as a
categorical variable for evaluating the overall motion
bed rank eect and as a continuous variable for
evaluating a trend eect. e MIXED procedure in
SAS Version 9.1.3 (SAS Institute, 2006) was used to
account for the correlation of mattresses and nights
within participants (Brown & Prescott, 2006; Senn,
2002). Statistical signicance refers to p < 0.05, and all
p-values are two-sided.
We used a similar regression modeling approach
to evaluate the eect of motion bed rank on eight
key diary outcomes, except in these models we
also adjusted for the participant’s age, gender, BMI,
ordinal day on bed, and time spent in bed (also
referred to as time devoted to sleep). e diary
outcomes are as follows: self-reported sleep time,
number of overnight awakenings, minutes awake
overnight, sleep quality, level of restedness at start
of day (also referred to as well-restedness at start of
day), pain upon waking (also referred to as morning
pain), daytime sleepiness, and daytime energy. Sleep
quality, restedness, morning pain, daytime energy,
and daytime sleepiness are self-reports using a
7-point Likert-type scale from least (1) to greatest
(7). Minutes awake were categorized and reported
using an ordinal scale (1=0 minutes, 2=1-15 minutes,
3=16-30 minutes, 4=31-45 minutes, 5=45-60 minutes,
6=60+ minutes). Self-reported sleep time was
recorded in minutes.
We evaluated the overall eect of motion bed rank
and the trend across bed rank on each of the diary-
reported sleep outcomes. We also performed pairwise
comparisons among motion bed ranks (best motion
bed vs. second-best bed; best motion bed vs. average
of all others) and estimated model-adjusted means
(also known as least square means) within each motion
bed rank category.
An example of the linear regression equation for one
sleep outcome, self-reported number of awakenings,
is as follows:
Number of Awakenings
ijk
= β
0
+ (β
1
x Motion Bed Rank
ij
)
+ (β
2
x Ordinal Day on Bed
ijk
)
+ (β
3
x Time in Bed
ijk
) + (β
4
x Age
i
)
+ (β
5
x Gender
i
) + (β
6
x BMI
i
)
+ (β
7
x Study Period 1
ijk
)
+ …+ (β
12
x Study Period 6
ijk
) ,
where β
0
— β
12
are the regression coecients to be
estimated. e response and independent variables are
dened as follows:
• Self-reported number of overnight awakenings
(response variable measured by the diary, for
subject i, bed j, night k).
• Motion bed rank (coded 1–7 for subject i, bed j,
modeled as categorical or continuous, depending on
whether the hypothesis is to evaluate an overall
eect of motion bed rank or to evaluate trend across
motion bed rank).
• Ordinal day on bed (this measures the acclimation
eect, coded 1 to number of days on bed, modeled
as continuous, for subject i, bed j, night k).
• Time in bed (amount of time devoted to sleep, in
minutes, as determined by actigraphy, modeled as
continuous, for subject i, bed j, night k).
• Age (in years at entrance into study, modeled as
continuous, for subject i).
• Gender (coded 1 for males, 0 for females, for subject i).
• BMI (body mass index, or weight/height
2
, modeled as
continuous, for subject i).
• Study period (Latin square crossover design variable,
coded 1–7, modeled as categorical, for subject i,
bed j).
Choosing the Best Mattress 5
Table 2 presents descriptive statistics for the response
variables used in the regression analyses and
describes the nature of the variables.
In addition to ranking the beds by overnight motion,
we also ranked the beds within subject by their
average self-reported sleep quality on each bed
(averaged over the nights that the subject slept on
the bed). Higher values of self-reported sleep quality
indicate better sleep on a 7-point scale. Based on this
ranking, a participant’s best sleep quality bed is the
mattress with the highest average sleep quality score.
We carried out a similar set of regression analyses as
previous, replacing motion bed rank with a ranking
based on self-reported sleep quality. Sleep quality
bed rank is modeled as continuous (one regression
coecient) for evaluating trend, and modeled as
categorical (six regression coecients) for all other
hypotheses.
Participants’ Self-Selection of “Best” Mattress
In the nal part of this analysis, we turned our
attention to whether the participant’s showroom
“test drive” provided a means of choosing a mattress
that predicts best sleep for that individual. More
specically, we evaluated whether the mattress
that individuals had said they would choose for
themselves agreed with their optimal mattress as
determined by actigraphy and separately by reported
sleep quality. Self-reported sleep quality is based on
diary reports of sleep quality during his or her in-
home testing.
To accomplish this, we estimated the kappa measure
of agreement
2
(Agresti, 2002) between the self-
selected bed and the actigraphically determined
best bed for each individual and also between the
self-selected bed and self-reported best sleep quality
bed for each individual. We also estimated the kappa
measure of agreement between the self-selected
bed and the top 3 best motion beds, based on the
observed vs. expected percentage of participants
for which the self-selected bed is among the motion
bed ranks of 1, 2, or 3 for the participant. Finally, we
estimated the mean and median motion bed rank
associated with an individual’s self-selected bed.
Table 2. Descriptive statistics for sleep variables
Variable (type and measure) N
Number
missing Minimum Mean Maximum
Actigraph
Overnight motion in minutes per night (continuous variable
normalized to 8 hours) 16,366 0 0 61.44 316.31
Time in bed devoted to sleep (continuous variable in minutes) 16,366 0 32.00 447.29 1,069.00
Diary
Self-reported sleep time (continuous variable in minutes) 15,941 425 0 433.78 888.00
Number of awakenings (continuous) 14,912 1,454 0 0.95 12.00
Minutes awake overnight (categorized from 1 to 6 in 15-minute
increments, where 1=0 minutes and 6=60+ minutes) 15,059 1,307 1.00 1.78 6.00
Sleep quality (categorical: 1=Not at all; 7=Very good sleep) 16,315 51 1.00 5.08 7.00
Well-restedness in AM (categorical: 1=Not at all; 7=Very well rested) 16,315 51 1.00 4.87 7.00
Morning pain (any type) (categorical: 1=None; 7=Worst imaginable) 16,282 84 1.00 1.56 7.00
Daytime sleepiness (categorical: 1=Not at all; 7=Very sleepy) 15,116 1,250
a
1.00 2.83 7.00
Daytime energy (categorical: 1=Not at all; 7=Very energetic) 15,116 1,250
a
1.00 4.86 7.00
a
Missing values for Daytime Sleepiness and Daytime Energy indicate lack of an evening diary report on the calendar day immediately following the previous night’s
actigraph data.
2
e kappa statistic measures the extent of agreement between two
raters beyond what would be expected by chance alone.
6 Hogan et al., 2011 RTI Press
As in our prior report, the analyses focused on the
128 participants who successfully completed the study
(Krystal et al., 2011). e previous report provides
complete details on our approach to including or
excluding individual observations for analysis.
Results
Actigraphically Determined Bed Ranking
(Motion Bed Rank)
e best motion bed was relatively evenly distributed
across mattresses in our sample. Figure 1 depicts
the frequency distribution of the actigraphically
determined best bed across mattresses. Table 3
presents the results of the regression modeling
(SAS MIXED procedure). e rst row of that table
determines the extent to which actigraph-measured
motion per night is lower on the best motion bed
compared to other beds. e results in the columns
labeled Bed Rank 1 through Bed Rank 7 report the
(model-adjusted) means for actigraphic and diary
measures. e Bed Rank 1 column indicates that the
mean overnight minutes of motion per 8 hours was
slightly more than 54 minutes on the actigraphically
determined best bed (Bed Rank 1). e Bed Rank
7 column reports that on average, actigraphic
measurement found nearly 69 minutes of overnight
motion on the worst bed (Bed Rank 7). is is a
dierence of 15 minutes and is statistically signicant
(p=0.0001).
In Table 3, the rightmost column indicates that the
best motion bed on average is associated with 3.26
fewer minutes of motion than the second-best bed.
is dierence is statistically signicant (p = 0.0001).
is column also reports that the best bed is
associated with 8.3 fewer minutes of motion than
the average of all other beds in the motion ranking
(second through seventh), and this, too, is statistically
signicant (p=0.0001).
Although the dierences in total motion are
numerically small, analysis of the eect of motion bed
rank on diary outcomes indicates that the bed with
lowest motion was signicantly associated with better
sleep quality, better feeling of restedness at the start
of the day, improved daytime energy, fewer nighttime
awakenings, and fewer minutes awake.
Self-reported sleep quality was signicantly
improved on the bed with the least overnight motion
compared with the bed with the second-lowest
motion (p=0.0048) and with the average of all other
beds (p=0.0010). In addition, sleep quality decreased
linearly with bed rank (p=0.0001). Average scores for
sleep quality ranged from 5.13 on the best motion
bed, to 5.05 on the second-best bed, to 4.99 on the
worst bed, out of a Likert-type scale of 1 (not at all
good) to 7 (very good).
Self-reported level of restedness was signicantly
improved on the best motion bed compared with the
average of all other beds (p=0.0193), and restedness
also decreased linearly with bed rank (p = 0.0078).
Average scores for restedness ranged from 4.91 and
4.87 on the best and second-best motion beds to 4.81
on the worst bed, out of a Likert-type scale of 1 (not
at all rested) to 7 (very well rested).
Self-reported daytime energy increased linearly
with bed rank, such that daytime energy tended to
increase in beds ranked higher on actigraphic sleep
(p=0.0016). However, daytime energy on the best
motion bed was only marginally increased when
compared to the average of all other beds (p=0.0574).
e average score for daytime energy ranged from
4.90 and 4.88 on the best and second-best motion
beds to 4.81 on the worst bed, out of a Likert-type
scale of 1 (not at all energetic) to 7 (very energetic).
Figure 1. Frequency distribution of best motion bed
15
18
20
19
15
21
20
11.7%
14.0%
15.6%
14.8%
11.7%
16.4%
15.6%
7654321
Study Beds
Number and Percent of Participants
0
5
10
15
20
25
Choosing the Best Mattress 7
Self-reported number of nighttime awakenings
were signicantly reduced on the best motion bed
compared with second best (p=0.0039) and compared
with the average of all other beds (p=0.0015). e
reduction in awakenings were also linearly related to
motion bed rank (p<0.0001). e average number of
nighttime awakenings ranged from 0.91 and 0.99 on
the best and second-best motion beds to 1.05 on the
worst bed.
Self-reported number of minutes awake overnight
was also signicantly reduced on the best motion bed
compared to the average of all other beds (p=0.0002),
and the reduction in minutes awake overnight was
also linearly related to motion bed rank (p<0.0001).
e average number of minutes awake overnight
(categorized in 15-minute increments from 1=0
minutes, 2=1-15 minutes,…, 6=60+ minutes) ranged
from 1.72 and 1.76 on the best and second-best
motion beds to 1.88 on the worst bed.
Table 3. Effect of motion bed rank on sleep outcomes
Sleep
Response
Variable
Model-Adjusted
a
Mean (SE)
Overall
Bed Rank
p-value
Rank 1 vs. 2
b
Rank 1 vs. All Others
c
Linear Trend
d
Bed
Rank 1
Bed
Rank 2
Bed
Rank 3
Bed
Rank 4
Bed
Rank 5
Bed
Rank 6
Bed
Rank 7
Actigraph
Overnight minutes
of motion per
8 hours
e
54.35
(1.63)
57.61
(1.63)
59.62
(1.63)
61.51
(1.63)
63.42
(1.63)
65.21
(1.63)
68.53
(1.63)
0.0001
-3.26 (0.38) p=0.0001
-8.30 (0.29) p=0.0001
2.20 (0.05) p=0.0001
Diary (Self-Report)
Sleep time
(minutes)
435.03
(2.33)
435.01
(2.33)
433.05
(2.32)
433.65
(2.33)
432.01
(2.32)
432.69
(2.33)
433.59
(2.33)
NS
NS
NS
-0.36 (0.19) p=0.0595
Number of
awakenings
0.9054
(0.0702)
0.9886
(0.0702)
0.9268
(0.0700)
0.9246
(0.0701)
0.9725
(0.0700)
0.9921
(0.0701)
1.0493
(0.0701)
0.0001
-0.08 (0.03) p=0.0039
-0.07 (0.02) p=0.0015
0.0175 (0.0039) p=0.0000
Minutes awake
overnight
1.72
(0.05)
1.76
(0.05)
1.73
(0.05)
1.79
(0.05)
1.83
(0.05)
1.81
(0.05)
1.88
(0.05)
0.0001
NS
-0.08 (0.02) p=0.0002
0.025 (0.004) p=0.0000
Sleep quality
5.13
(0.07)
5.05
(0.07)
5.08
(0.07)
5.10
(0.07)
5.07
(0.07)
5.07
(0.07)
4.99
(0.07)
0.0001
0.08 (0.03) p=0.0048
0.07 (0.02) p=0.0010
-0.015 (0.004) p=0.0001
Well-restedness
4.91
(0.07)
4.87
(0.07)
4.88
(0.07)
4.88
(0.07)
4.87
(0.07)
4.88
(0.07)
4.81
(0.07)
0.0347
NS
0.05 (0.02) p=0.0193
-0.01(0.004) p=0.0078
Morning pain
severity
1.53
(0.06)
1.56
(0.06)
1.51
(0.06)
1.55
(0.06)
1.60
(0.06)
1.53
(0.06)
1.56
(0.06)
0.0263
NS
NS
NS
Daytime sleepiness
2.78
(0.08)
2.82
(0.08)
2.81
(0.08)
2.85
(0.08)
2.76
(0.08)
2.77
(0.08)
2.90
(0.08)
0.0002
NS
NS
NS
Daytime energy
4.90
(0.08)
4.88
(0.08)
4.85
(0.08)
4.93
(0.08)
4.90
(0.08)
4.84
(0.08)
4.81
(0.08)
0.0001
NS
0.04 (0.02) p=0.0574
-0.01 (0.003) p=0.0016
BMI = body mass index; NS = not statistically significant (p > 0.05); SE = standard error.
a
Diary means within levels of motion bed rank are adjusted for the following model covariates: Ordinal Day in Bed, Time in Bed, Age, Gender, BMI, and Study Period.
b
Bed Rank 1 minus Bed Rank 2: Estimated difference, SE, and p-value (2-sided).
c
Bed Rank 1 minus average of Bed Ranks 2–7: Estimated difference, SE, and p-value (2-sided).
d
Linear trend across Bed Ranks (1=Best, 7=Worst): Estimated slope, SE, and p-value (2-sided).
e
Means within levels of motion bed rank are not adjusted for any covariates. They are included as descriptive information.
NOTE: Bed Rank 1 indicates least overnight motion measured by actigraphy, and Bed Rank 7 indicates greatest motion measured by actigraphy. All analyses carried
out using the SAS MIXED procedure.
8 Hogan et al., 2011 RTI Press
Sleep Quality Bed Ranking
Table 4 reports the results of a similar analysis
identifying the “best bed” as determined by the diary
measure of overnight sleep quality. In the rst row of
the table, the columns labeled Bed Rank 1 through
Bed Rank 7 report self-report sleep quality means
along the Likert scale where 1 is worst sleep quality
and 7 is best. On average, all bed ranks rate above
the midpoint (4), with the best sleep quality bed on
average rated 5.6 on the scale and the second-best
rated 5.41. e lowest rating bed had a sleep quality
score of 4.32. e far-right column indicates an
average increase of 0.19 (3.5 percent improvement) in
sleep quality between the best and second-best sleep
quality bed (p<0.0001). is column also indicates an
increase of 0.65 (13 percent improvement) in sleep
quality on the best bed compared with the average
of the remaining bed ranks (p<0.0001). Sleep quality
decreased linearly with bed rank (p<0.0001).
Most importantly, there was a signicant relationship
between sleep quality bed rank and actigraphic
Table 4. Effect of sleep quality bed rank on actigraphic motion and other sleep outcomess
Sleep
Response
Variable
Model-Adjusted
a
Mean (SE)
Overall
Bed Rank
p-value
Rank 1 vs. 2
b
Rank 1 vs. All Others
c
Linear Trend
d
Bed
Rank 1
Bed
Rank 2
Bed
Rank 3
Bed
Rank 4
Bed
Rank 5
Bed
Rank 6
Bed
Rank 7
Sleep Quality
e
5.60
(0.07)
5.41
(0.07)
5.24
(0.07)
5.10
(0.07)
4.93
(0.07)
4.74
(0.07)
4.32
(0.07)
<.0001
0.19 (0.03) p<0.0001
0.65 (0.02) p<0.0001
-0.20 (0.0050) p<0.0001
Actigraph
Overnight minutes
of motion per
8 hours
61.53
(1.64)
61.97
(1.64)
62.54
(1.64)
62.55
(1.64)
62.21
(1.64)
63.12
(1.64)
62.62
(1.64)
0.0233
-0.43 (0.46) p=0.3501
-0.97 (0.34) p=0.0055
0.19 (0.06) p=0.0023
Diary (Self-Report)
Sleep time
(minutes)
434.41
(2.31)
436.55
(2.33)
434.69
(2.32)
432.56
(2.33)
433.08
(2.34)
430.98
(2.33)
432.36
(2.35)
0.0021
-2.14 (1.37) p=0.1210
1.04 (1.03) p=0.3156
-0.68 (0.18) p=0.0003
Number of
awakenings
0.80
(0.07)
0.87
(0.07)
0.92
(0.07)
0.96
(0.07)
0.95
(0.07)
1.08
(0.07)
1.23
(0.07)
<.0001
-0.07 (0.02) p=0.0168
-0.20 (0.02) p<0.0001
0.06 (0.0038) p=<0.0001
Minutes awake
overnight
1.66
(0.05)
1.68
(0.05)
1.76
(0.05)
1.78
(0.05)
1.80
(0.05)
1.90
(0.05)
1.98
(0.05)
<.0001
-0.02 (0.02) p=0.5018
-0.16 (0.02) p<0.0001
0.05 (0.0037) p<0.0001
Well-restedness
5.27
(0.07)
5.15
(0.07)
5.02
(0.07)
4.89
(0.07)
4.76
(0.07)
4.57
(0.07)
4.32
(0.07)
<.0001
0.12 (0.02) p<0.0001
0.49 (0.01) p<0.0001
-0.15 (0.0034) p<0.0001
Morning pain
severity
1.34
(0.06)
1.42
(0.06)
1.51
(0.06)
1.48
(0.06)
1.59
(0.06)
1.70
(0.06)
1.86
(0.06)
<.0001
-0.08 (0.02) p=0.0029
-0.25 (0.02) p<0.0001
0.08 (0.0036) p<0.0001
Daytime
sleepiness
2.64
(0.08)
2.71
(0.08)
2.76
(0.08)
2.83
(0.08)
2.82
(0.08)
2.96
(0.08)
3.00
(0.08)
<.0001
-0.07 (0.03) p=0.0401
-0.20 (0.02) p<0.0001
0.06 (0.0045) p<0.0001
Daytime energy
5.04
(0.08)
5.02
(0.08)
4.93
(0.08)
4.89
(0.08)
4.79
(0.08)
4.76
(0.08)
4.63
(0.08)
<.0001
0.02 (0.02) p=0.4659
0.20 (0.01) p<0.0001
-0.07 (0.0033) p<0.0001
BMI = body mass index; NS = not statistically significant (p > 0.05); SE = standard error.
a
Means within levels of Sleep Quality Bed Rank are adjusted for the following model covariates: Ordinal Day in Bed, Time in Bed, Age, Gender, BMI, and Study Period.
b
Bed Rank 1 minus Bed Rank 2: Estimated difference, SE, and p-value (2-sided).
c
Bed Rank 1 minus average of Bed Ranks 2–7: Estimated difference, SE, and p-value (2-sided).
d
Linear trend across Bed Ranks (1=Best, 7=Worst): Estimated slope, SE, and p-value (2-sided).
e
Means not adjusted for any covariates. They are included as descriptive information.
NOTE: Bed Rank 1 indicates highest sleep quality measured by diary, and Bed Rank 7 indicates lowest sleep quality measured by diary. All analyses carried out in the
SAS MIXED procedure.
Choosing the Best Mattress 9
motion (p=0.0233) as well as many diary outcomes,
most notably number of awakenings (p<0.0001),
number of minutes awake overnight (categorized)
(p<0.0001), well-restedness (p<0.0001), morning pain
severity (p<0.0001), daytime sleepiness (p<0.0001),
and daytime energy (p<0.0001). Aer adjusting
for covariates, actigraphic motion minutes was
signicantly reduced in the best sleep quality bed
compared to the average of all other bed ranks, and
motion increased linearly with bed rank. e beds
ranked best and second-best for self-reported sleep
quality are also the two beds with the lowest level of
actigraphically measured motion.
e number of awakenings and minutes awake
overnight were both signicantly reduced in the
best sleep quality bed compared to the average of
all other bed ranks, with each increasing linearly
with bed rank. e number of awakenings was
also signicantly reduced in the best sleep quality
bed compared to second best. Well-restedness and
daytime energy were both signicantly improved in
the best sleep quality bed compared to the average
of all other bed ranks, with each decreasing linearly
with bed rank. Well-restedness was also signicantly
improved in the best sleep quality bed compared
to second best. Finally, morning pain severity and
daytime sleepiness were both signicantly reduced
in the best sleep quality bed compared to the second
best bed and compared to the average of all other bed
ranks, with each increasing linearly with bed rank.
Self-Selected Versus Actigraphically
Determined Best Bed
We conducted a series of analyses to determine
whether the mattress that individuals indicated they
would choose for themselves was predictive of their
optimal mattress (or mattresses) as determined
by either actigraphy or self-reported sleep quality.
In other words, can people predict their optimal
mattress from a typical in-store experience? Results
indicate that standard showroom testing does not
lead individuals to select the bed that will provide
their best sleep as measured by either self-reported
sleep quality or actigraphic measurement over an
extended period of time. Although the self-selected
bed varied in our sample (see Figure 1), it was
associated with a median motion bed rank of only
4 out of 7 (i.e., mid-rank), and the same was true
for self-reported sleep quality (Table 5). Consistent
with this, the kappa measure of agreement between
the best motion bed and self-select bed was not
signicantly dierent from 0, indicating no additional
agreement than what would be expected by chance
alone (Table 6). Agreement between self-select bed
and the top 3 best motion beds (motion bed rank of
1, 2, or 3) was also not signicantly dierent from
0 (Table 7), as the percentage of participants whose
self-select bed was among the top 3 motion best
beds was only 38 percent (vs. 43 percent expected
by chance alone). is suggests that the customary
showroom “test drive” in fact oen leads consumers
to suboptimal mattress selection.
Table 5. Average motion bed rank and self-reported
sleep quality bed rank of self-selected bed
Variable Used
to Determine
Bed Rank Statistic
Estimated
Bed Rank
(1–7)
Lower 95%
Limit
Upper 95%
Limit
Actigraphic
Motion
a
(in minutes)
Average 4.172 3.826 4.518
Median 4.000 4.000 5.000
Self-Reported
Sleep Quality
(coded 1–7)
Average 3.875 3.525 4.225
Median 4.000 3.000 4.000
a
Actigraphic motion is number of minutes of recorded overnight motion,
normalized to an 8-hour night.
Table 6. Agreement between self-selected bed and
best motion bed
Methods
Compared
Kappa
Estimate
Lower 95%
Limit
Upper 95%
Limit
Self-Select vs. Best
Motion
a
Bed
-0.0114 -0.0795 0.0567
a
Best motion bed is the mattress firmness with the smallest average overnight
motion minutes per 8 hours.
Table 7. Agreement between self-selected bed and
top 3 best motion beds
Methods
Compared
Kappa
Estimate
Observed
Percentage
a
Expected
Percentage
b
P-value
H
0
: Kappa=0
Self-Select vs.
Top 3 Best
Motion
c
Beds
-0.0801 38.28% 42.86% 0.2904
a
Observed percentage of people for which self-select bed is among the top 3
best motion beds (motion bed ranks 1, 2, or 3).
b
Expected percentage of people for which self-select bed is among the top 3
best motion beds (motion bed ranks 1, 2, or 3).
c
The top 3 best motion beds are the three mattress firmness levels with the
smallest average overnight motion minutes per 8 hours.
10 Hogan et al., 2011 RTI Press
Discussion
Our prior report documented that mattress rmness
has signicant eects on sleep and daytime function
(Krystal et al., 2011). Very low and very high levels of
rmness tended to be associated with relatively worse
sleep, greater morning pain, and poorer daytime
function.
Our results build on those observations. We provide
here six main conclusions from our blinded,
controlled study:
1. Individuals dier as to the degree of mattress
rmness that is associated with their best sleep.
2. Actigraphic activity level has signicant utility
for identifying the best mattress for an individual
(Table 3).
3. e best bed as measured by actigraphic motion
signicantly minimizes self-reported number
of overnight awakenings and minutes awake
overnight, and signicantly maximizes self-
reported sleep quality and well-restedness in the
morning (Table 3).
4. e best bed as determined by self-reported sleep
quality in the morning signicantly minimizes
actigraphic motion, self reported number of
awakenings, minutes awake overnight, and
morning pain severity, and maximizes well-
restedness, daytime sleepiness, and daytime energy
(Table 4).
5. People are essentially le to chance when trying to
select a mattress using the generally recommended
“test drive” on the showroom oor (Tables 5–7).
6. e actigraphically best mattress was well
distributed among the sample of participants and
across the seven mattresses studied.
ese data indicate that activity level is not a
reection of mattress rmness; rather, it provides
an indication of the mattress that yields optimal
sleep for an individual. is is consistent with our
previously reported nding that activity level was
not signicantly correlated with mattress rmness
(Krystal et al., 2011). us, our ndings imply that
improving sleep in many individuals by improving
mattress t is quite possible. Our results also suggest
that actigraphically determined activity level may
have some utility in this regard.
ese observations suggest that improving mattress
t may improve sleep. Pain, daytime sleepiness, and
energy level appear to be aected more by mattress
rmness than by the degree to which rmness is
suitable for an individual. is may reect the fact
that we excluded individuals with pain, insomnia,
or daytime sleepiness from this study. Assessing the
eects of mattress t on these measures in studies that
include such individuals is an important question for
future research.
e mattress that individuals chose as optimal before
the randomized, controlled phase of the study did
not predict either the actigraphically determined
best mattress or the best mattress as determined
by reported sleep quality. is nding raises the
possibility that the ordinary showroom experience
does not lead individuals to select the mattress that
results in best sleep over a more extended period.
is nding should inspire study into ways in
which consumers can be better equipped to identify
mattresses that lead to optimal sleep, and the health
benets that come from better rest.
In summary, our study indicates that optimal
mattress rmness varies among individuals and is
reected, at least to a degree, by actigraphic activity
level. When allowed to test mattresses in a typical
showroom experience, individuals appear to choose
mattresses that do not optimize their sleep. Given
that this “test drive” approach is commonplace, it
would seem that most of the general public may
be sleeping on mattresses improperly suited to the
individual owners. is would help explain to some
extent why so many Americans are not enjoying the
health benets that come from optimal rest. Sleep
science could assist bedding retailers in improving
their customers’ sleep outcomes by developing better
in-store methods of aiding in the mattress selection
process.
Choosing the Best Mattress 11
References
Hanel, S. E., Dartman, T., & Shishoo, R. (1996).
Measuring methods for comfort rating of seats
and beds. International Journal of Industrial
Ergonomics, 20, 163–172.
Krystal, A. Edinger, J., Bieler, G., Mladsi, S., &
Hogan, S. (2011). Mattress matters: A randomized,
controlled, Latin-square crossover study of the eects
of mattress support on sleep, pain, and daytime
functioning. Research Triangle Park, NC: RTI Press.
Laird, N. M., & Ware, J. H. (1982). Random eects
models for longitudinal data. Biometrics, 38,
963–974.
Lindsay, J. (1993). Models for repeated measurements.
Oxford: Oxford University Press.
Littner, M. Kushida, C. A., Anderson, W. M.,
Bailey, D., Berry, R. B., Davila, D. G., Hirshkowitz,
M., Kapen, S., Kramer, M., Loube, D., Wise, M.,
& Johnson, S. F. (2003). Practice parameters for
the role of actigraphy in the study of sleep and
circadian rhythms: An update for 2002. Sleep, 26,
337–341.
López-Torres, M., Porcar, R., Solaz, J., & Romero, T.
(2008). Objective rmness, average pressure and
subjective perception in mattresses for the elderly.
Applied Ergonomics, 39, 123–130.
Morgenthaler, T., Alessi, C., Friedman, L., Owens, J.,
Kapur, V., Boehlecke, B., Brown, T., Chesson, A. Jr.,
Coleman, J., Lee-Chiong, T., Pancer, J., & Swick,
T. J. (2007). Practice parameters for the use of
actigraphy in the assessment of sleep and sleep
disorders: An update for 2007. Sleep, 30(4),
519–529.
National Institutes of Health (NIH), National Institute
of Neurological Disorders and Stroke. (2003a).
Brain basics: Understanding sleep. Bethesda, MD:
National Institute of Neurological Disorders and
Stroke.
National Institutes of Health (NIH), National Center
for Sleep Disorders Research. (2003b). National
sleep disorders research plan. NIH Report No.
03-5209. Bethesda, MD: National Institute of
Neurological Disorders and Stroke.
Agresti, A. (2002). Categorical data analysis, 2nd ed.
New York: John Wiley and Sons.
Alapin, I., Fichten, C. S., Libman, E., Creti, L., Bailes,
S., & Wright, J. (2001). How is good and poor
sleep in older adults and college students related
to daytime sleepiness, fatigue and ability to
concentrate? Journal of Psychosomatic Research,
49(5), 381–390.
Ancoli-Israel, S., Cole, R., Alessi, C., Chambers,
M., Moorcro, & Pollak, C. P. (2003). e role
of actigraphy in the study of sleep and circadian
rhythms. Sleep, 3, 342–392.
Better Sleep Council. (2008a). Consumer attitudes
about mattresses are evolving, but purchasing
behavior is slow to change. Sleep Savvy, April.
Better Sleep Council. (2008b). Mattress buying guide.
Available at http://www.bettersleep.org/Ease
/shop.asp
Brown, H., & Prescott, R. (2006). Applied mixed
models in medicine, 2nd ed. New York: John Wiley
and Sons.
Centers for Disease Control and Prevention (CDC).
(2007). Sleep and sleep disorders: Sleep hygiene
tips. Available at http://www.cdc.gov/sleep
/hygiene.htm
Consumer Reports. (2005). How to buy a mattress
without losing sleep. June, 12–17.
Diggle, P. J., Liang, K. Y., & Zeger, S. L. (1994).
Analysis of longitudinal data. Oxford: Oxford
University Press.
Elmenhorst, E. M., Elmenhorst, D., Luks, N., Maass,
H., Vejvoda, M., & Samel, A. (2008). Partial sleep
deprivation: Impact on the architecture and
quality of sleep. Sleep Medicine, 8, 840–850.
Hamilton, N. A., Gallagher, M. W., Preacher, K. J.,
Stevens, N., Nelson, C. A., Karlson, C., &
McCurdy, D. (2007). Insomnia and well being.
Journal of Consulting & Clinical Psychology, 75,
939–946.
12 Hogan et al., 2011 RTI Press
Roberts, R. E., Roberts, C. R., & Duong, H. T. (2008).
Chronic insomnia and its negative consequences
for health and functioning of adolescents: A
12-month prospective study. Journal of Adolescent
Health, 42, 294–302.
SAS Institute Inc. (2006). SAS/STAT user’s guide,
Version 9.1.3. Cary, NC: SAS Institute Inc.
Senn, S. (2002). Cross-over trials in clinical research,
2nd ed. New York: John Wiley and Sons.
Van Dongen, H. P., Maislin, G., Mullington, J. M.,
& Dinges, D. F. (2003). e cumulative cost of
additional wakefulness: Dose-response eects on
neurobehavioral functions and sleep physiology
from chronic sleep restriction and total sleep
deprivation. Sleep, 2, 117–126.
Acknowledgments
Sleep to Live Institute of Joplin, Missouri, sponsored this research. Sleep to Live
Institute did not exercise editorial control over the analysis or reporting of the
results. We recognize the contributions of Michael S.S. Lawrence, Je Barghout,
and Scott Mladsi in the design and implementation of this experiment.
RTI International is an independent, nonprot research organization dedicated
to improving the human condition by turning knowledge into practice. RTI
oers innovative research and technical solutions to governments and businesses
worldwide in the areas of health and pharmaceuticals, education and training,
surveys and statistics, advanced technology, international development,
economic and social policy, energy and the environment, and laboratory and
chemistry services.
e RTI Press complements traditional publication outlets by providing another
way for RTI researchers to disseminate the knowledge they generate. is PDF
document is oered as a public service of RTI International.
www.rti.org/rtipress RTI Press publication RR-0016-1108