Number of People in the United States Experiencing Ambulatory and Independent Living Difficulties

Abstract

Estimating the characteristics of the “disabled” population is necessary for some governments and of interest to health researchers concerned with producing disability prevalence rates. Because generating easy-to-understand estimates of disability in the population is important, this article provides US population estimates for two disability-related measures by using the 2009 to 2011 American Community Survey Public Use Microdata Sample file. The number of people who have “independent living” and “ambulatory” difficulties is calculated from a sample of 9,204,437 (representing > 309 million people). The percentage for “disabled” is found to vary by: racial-ethnic category, sex, age, citizenship status, educational attainment, and state-level regions divided by weather.

Differently-abled (commonly referred to as “disabled”) (Wendell, 1989) United States (US) residents are an underserved community whose physiological conditions present unique social and physical struggles—arising from the mismatch between their personal and social expectations and available resources in the environment. Some “able-bodied” people, by virtue of being able-bodied, lack an understanding of how small physical obstacles (e.g., lack of pedestrian ramps at street crossways) can present a noticeable challenge for the differently-abled (Clarke & George, 2005). Even though it may be that definitions of “disability” vary by time and geography (Baynton, 2013), in a classical disablement process framework, having difficulties with walking up stairs or going shopping could be said to represent pre-clinical physiological limitations which could lead to the formation of clinical disability (Verbrugee & Jette, 1994).

To be sure, the sociologically produced drive for physical sameness creates interest in understanding the characteristics of the differently-abled population (Munyi, 2012). A view is taken throughout this paper that disability is a problem which can only be situated in the interaction between disabled individuals and their social environment. From this perspective, addressing challenges faced by disabled individuals requires a psychosocial and physical adjustment on the part of society (Karger & Rose, 2013). This paper is motivated by the perspective that resources for coping with physical limitations are non-randomly (i.e., unevenly) distributed through a stratifying social system. Developing estimates on the prevalence of disability is intended to help federal, state, county, and local governments assess the impact of policies aimed at reducing discrimination against the differently abled and to improve their participation in community activities (Brault, 2009).

Some evidence exists that the unjust stigmatization of the differently abled (Bourke & Waite, 2013) can lead some of them to have lower levels of social participation (McPhedran, 2012). This point matters because arguments have been made that exposures over the life course can influence health outcomes in adulthood through a cumulative process (Blane, 1999). This approach views cumulative exposures to adverse social circumstances (e.g., poverty, discrimination) as being important risk factors for disease (Shoham et al., 2008). From this perspective, it may be argued that adverse health outcomes linked with the disabled status can be aggravated by the presence of other known risk factors. If you note that socioeconomic and racial/ethnic disparities in health are well documented (Arauz Boudreau, Kurowski, Gonzalez, Dimond, & Oreskovic, 2013), then you may be able to see why, for example, the effects on health (e.g., formation of excess adipose tissue) from not being able to walk may be compounded upon by a person’s financial status (i.e., being “poor”) and/or place of residence (e.g., living in a dangerous neighborhood where healthy food options are scarce).

The specific aim of this paper is to highlight how population estimates of disability prevalence vary in their precision in unique ways (Siordia, 2013b). Because data from the American Community Survey (ACS) may be argued to be the gold standard for producing reliable measures of disability in the general US population, it has been deemed critical to the disability community. For example, local governments (e.g., states) use ACS data to decide how to distribute funds to local agencies for food, health care, and legal services for individuals with disabilities. If the best available data has the potential to under- or over-estimate the prevalence of disability as a function of demographic characteristics, then there is a possibility that “multi-disadvantaged” (e.g., disabled racial/ethnic minorities with moderate levels of educational attainment) groups may not be receiving the appropriate level of resources.

Previous work has shown how data may be challenged if “selection bias” (Kleinbaum, Morgenstern, & Kupper, 1981), the selection of study participants by a third unmeasured variable believed to be associated with exposure and outcome, is suspect (Strandhagen et al., 2010). Please note that although ACS may have some limitations commonly found in population based survey studies, an argument is not being made here that ACS data contains selection bias. In 2011 and at the national level, the ACS had a coverage rate of 98.6% and a “response rate” of 97.6%—numbers which may be argued to indicate there is a low probability of selection bias. Instead, the main point of this discussion is to advise researchers interested in estimating disability prevalence in the US population to consider how precision of the estimates varies from group to group. Between group comparisons is a particularly important point when you consider that research on “health disparities” continues to grow (e.g., Pollack et al., 2013). A related issue has been raised by others on how temporal comparisons (i.e., time dependent) of socioeconomic disparities between cross-sectional surveys may be affected when presumed variations on the meaning of a socioeconomic status are unaccounted for in statistical modeling (Chen, Beckfield, Waterman, & Krieger, 2013).

Estimating the number of individuals within the US population who may be facing uncommon social and physical challenges is important for assisting public health efforts towards developing assistance measures. US government policies provide some protection for individuals with disability (Karger & Rose, 2013) and as a result, there is a mandate for US federal agencies collecting information on the US population to develop a count of individuals with disability. In particular, Title II of the Americans with Disabilities Act (ADA, 1990), requires government agencies to make services available to people with disabilities. In this study, two measures of disability are used to present a profile of the US population. These may be said to be related to the popularly conceptualized and measured activities of daily living (Katz, 1983).

Details of the survey questions used in the analysis are delineated below. The discussion here only focuses on the labels being used. Reported difficulties with “independent living” (e.g., ability to go shopping) are labeled as Outdoor Physical Mobility (Out_PM) and “ambulatory” (e.g., ability to walk) difficulties are referred to as Indoor Physical Mobility (In_PM). Both Out_PM and In_PM are treated as crude measures of potential capacity to function in the hypothetical tense (Glass, 1998). Because the items being used in the ACS to assess disability status are based on either self- or proxy-report (more explained below), they are referred to as “potential capacity”—in contrast, enacted mobility refers to actual performance (Siordia, 2013a). Out_PM and In_PM subjectively assess whether a person perceives an ability to perform a task. As a consequence, the labels being used here only signal “potential for outdoor physical mobility” with the Out_PM item and “potential for indoor physical mobility” with the In_PM item.

The report makes use of survey data derived from a US population-sample. A population-sample is a sub-set of individuals from a complete population (i.e., the population universe). The need to generalize findings from a sample, to the complete population from which they are drawn, necessitates the use of random selection. Because simple random selection is so infrequently available, the use of complex sampling techniques produces the need to use intricate weighting methodologies to derive population estimates from samples (Siordia & Lee, 2013). Since the sample is much smaller than the population/universe, each study subject represents a certain number of his/her peers. The use of “weights” amplifies sample numbers to derive estimates of the population. I now pause to clarify the sometimes not so obvious meaning of a key term: estimates. Producing an estimate of disability means math is being used with intelligently gathered information to produce an informed guess of the number of people with x-type of disability (Siordia, 2013b).

Inherent in this approach is the assumption that the informed guess (i.e., estimate) has a certain degree of uncertainty—a range of numbers where the “true” value is expected to exist. An estimate should not be seen as an inferior product with artificial significance. Instead, the reader should note that ACS is the most reliable source of information for estimating disability prevalence in the US population and that the estimates being presented here are amongst the most scientifically sophisticated measures available to non-government-affiliated researchers. The US has more than 310 million people spread over more than 3.8 million square miles. Trying to gather data on this population to describe their characteristics is incredibly challenging and critical for the advancement of democratic governance. A discussion on the variability of precision in disability estimates should not distract the reader from the fact that the ACS is one of the most sophisticated products created by the US Census Bureau. It is only because the ACS is so highly regarded, transparent to the public, and scientifically sound that a discussion on “estimate precision” is even possible—some small datasets purporting to estimate the prevalence of disability in the US would not stand the scrutiny being presented here and as applied to ACS microdata.

This paper introduces the reader to the idea that US population estimates of the disabled population vary in quality and as a function of person-characteristics and geographical location. Although technical in nature, the paper is meant to primarily provide detailed disability estimates by commonly used demographical characteristics. The specific aim of this paper is to provide US population estimates for Out_PM and In_PM. This aim is complimented by a discussion of various easy to understand “measures on uncertainty” in the estimations of Out_PM and In_PM for the US population during the 2009-2011 year-period.

Methods

Data

Disability in the US population is measured by taking a “randomly selected sample” from the population. As alluded to earlier, a sample is a sub-group derived from the population and randomly selected refers to a very important principle in inferential statistics. Randomly selected means that a person was selected from a population of individuals where every person had an equal chance for selection. This is important because subjects in the sample are used to infer the characteristics of the population they were drawn from—inferences are made by using “population weights” which assume a particular selection process. Because the ACS is not a simple random sample, population weights are variables said to account for the design of the study, the sampling error in the study—i.e., variation in the probability of being randomly selected at different points of the complex survey design (Siordia, 2014).

This study used data from the American Community Survey (ACS). In particular, the Public Use Microdata Sample (PUMS) 3-year file for 2009-2011 is used. Because ACS is collected yearly, the 3-year file being used here represents data collected over 36 months. The microdata are used internally by the US Census Bureau to produce estimates for geographies (e.g., counties) with as little as 20,000 people. Data from the PUMS files only allow geo-referencing to the Public Use Microdata Area (PUMA) geography—in 2013, the “5-year” ACS file (data from 60 months) will provide “disability” data all the way down to the block-group geography. Please note that data issues arising from the misapplication of disclosure-avoidance procedures (Alexander, Davern, & Stevenson, 2010) are not present in the microdata being used in this study. This is the first publication showing disability estimates for the population aged 65 and over where the PUMS data files are not believed to be contaminated by flawed disclosure-avoidance procedures. The estimates in this paper’s tables use a total of 9,204,437 actual survey participants. The use of population weights are then applied to the 9,204,437 individuals —actual number of people participating in the survey and referred to hereafter as the “unweighted sample.” When “weighted up” (using the PWGTP variable in the microdata), the unweighted sample is said to represent the US population, which for this time period is about 309,231,245 people (estimated total population of the US during the 2009-2011 period).

“Disability” Variables

Although there is information on six “disabilities” (label used by US Census Bureau), the current project only focuses on “independent living difficulty” and “ambulatory difficulty.” Readers should note that only one person per household participated in the survey and is technically referred to as the “reference person.” The reference person answered all questions for individuals in the household. Thus, if the reference person is responding to disability items for her/himself, then disability can be said to be self-reported. However, if the reference person is responding about the disability of others in the household, then disability can be said to be proxy-reported.

To assess independent living difficulty, survey respondents were asked about themselves and others in the household: “Because of a physical, mental, or emotional condition, does this person have difficulty doing errands alone such as visiting a doctor’s office or shopping?” (U.S. Census, 2011). Because of the way the questions are framed and the fact that individuals are only allowed to respond in a forced choice format (i.e., either yes or no), it may be argued that survey responses are approximate measures for outdoor physical mobility potential (i.e., Out_PM). To assess ambulatory difficulty, survey respondents were asked about themselves and others in the household: “Does this person have serious difficulty walking or climbing stairs?” (U.S. Census, 2011). This item may also be argued to be an approximate measure for indoor physical mobility potential (i.e., In_PM).

Elegant work has been undertaken to show that responding to these questions is primarily challenged by the need to determine whether or not the severity of a particular functional limitation warranted a positive response (Miller & DeMaio, 2006). For example, with regards to the In_PM, cognitive research showed that individuals considered physical limitation factors (e.g., pain, fatigue) when evaluating their ability to walk or climb stairs, but considered other domains (e.g., emotional status) when responding about others in the household. Becuase the In_PM question does not include the word “usually,” it may fail to prompt the respondent to think of long-term physical conditions. Because respondents are not asked about the use of assistive devices, some may see the use of a device as granting mobility and may thus fail to report any difficulty (Miller & DeMaio, 2006). With regards to Out_PM, responses to the question may vary because individuals may interpret “difficulty doing errands alone” differently. Some may interpret the question as probing for the person’s access to transportation resources, while others may interpret it as pertaining to mobility and mind capacity (Miller & DeMaio, 2006). In short, care should be given when comparing disability prevalence between studies using different forms of questions.

Demographic Variables

In_PM and Out_PM estimates are presented by various demographic factors. Because the US Census Bureau is forthcoming with how they collect information on the US population, documentation detailing the various labels under these racial-ethnic categories is readily available online. I made use of six race-ethnic categories: Non-Latino-White (NLW); Non-Latino-Black (NLB); Non-Latino-Other (NLO); Mexican-Latino (ML); and Non-Mexican-Latino (NML). The US Census Bureau conceptualizes race and ethnicity as a social construct—not a genetically determined and biologically defined phenotype.

Estimates are also provided for the following: males and females; three age groups (age ≤ 49; age 50-64; and age ≥ 65); three citizenship status categories (US born; naturalized; and non-citizen); educational attainment (≤ 1 year of college; ≥ Associates degree). To complement these popular measures, estimates by “warmer South” and “colder North” states were created. The “average annual extreme minimum temperature from 1976-2005” data from the United States Department of Agriculture (USDA) is used to qualitatively identify the states where about 50% of their geographical area experienced temperatures ≥ 10 °F (USDA, 2012). This arbitrary and visual approach renders the following as warmer South states: Alabama; Arizona; California; Florida; Georgia; Hawaii; Louisiana; Mississippi; New Mexico; Puerto Rico (not technically a “state”); South Carolina; and Texas. All the others make up the colder North states.

Statistical Approach

Because the US Census Bureau is so transparent about their methods for estimating population characteristics, I was able to make use of 80 person-weights (PWGTP1-PWGTP80 variables) provided in the PUMS files to estimate various measures of uncertainty surrounding the estimate. Replicate weights are used to calculate what is referred to as “direct standard errors”—measures of imprecision for the 80 possible estimates. I developed an algorithm, in SAS® 9.3, using the 80 person-weight by following instructions outlined in US Census Bureau publications (U.S. Census, 2009a, 2009b). The formula for calculating the “replicate weights standard errors” is as follows:

$$\mathit{SE}\left(X\right)=\sqrt{\frac{4}{80}\sum _{\mathit{r}=1}^{80}{\left(X_r-X\right)}^2}$$

After the replicate estimates X₁ through X₈₀ are computed, the standard error (SE) of X is estimated using the sum squared differences between each replicate estimate of X_r and the full sample estimate X (U.S. Census Bureau, 2009b). A single-person weight (i.e., PWGTP) to compute “weighed” versus “unweighted” estimates is used. What is hoped to be a more intuitive representation of variability of precision is the estimate presented by computing the Person Inflation Ratio (PIR), which is the average number of people being represented in weighted population by the unweighted counts. It is computed using the following formula: [weighted count ÷ weighted total population] (Siordia & Lee, 2013). An increase in PIR indicates that each individual within that group represents more of their counterparts in the population. When PIR increases, it indicates that the characteristics of fewer individuals may affect the estimation of disability prevalence.

ACS PUMS files allow public data users to calculate the margin of error (MOE) for each of the population estimates. To highlight how any estimates have a certain degree of uncertainty, I present MOEs for the estimates being produced. MOE is present when, for example, large variation between samples occurs—leading survey-based estimates to deviate from accurately approximating the true population value. These deviations from the true count can be roughly measured by computing the standard error (SE) of the estimate (Siordia & Lee, 2013). The Margin of Error (MOE) of the estimate with a 90% confidence is calculated as follows: [MOE=1.645*SE]. MOEs are then used to provide readers with the “upper” and “lower” 90% confidence limits around the disability estimates—the range in which the “true” number may lie with a 10% chance for making an error. Confidence limits are computed with the following equations: Lower Confidence Limit [LCL=(estimate – MOE)]; and Upper Confidence Limit [UCL=(estimate + MOE)]. A more easy to understand measure called Range of Uncertainty (RU) (Siordia & Lee, 2013) is also provided. RU compares the level of uncertainty in the disability estimate by using the following formula: {[(SE*3) ÷ X]*100}, where X is the disability estimate. By using this approach, readers can more straightforwardly interpret level of uncertainty as follows: as RU numbers increase, the level of imprecision in the estimate increases.

There is a very important point that has not been mentioned until now: In order to provide complete data banks, the US Census Bureau undertakes many widely accepted statistical procedures for insuring coherent responses and the reduction of missing items. Although the issue of “allocation” is explained in greater detail elsewhere (Siordia & Young, 2013), I will briefly mention that both probability (using statistical techniques) and non-probability (using explicit logic rules) based computer algorithms are used to assign responses to individuals missing a response to a disability survey-item or providing what is deemed an illogical response to a disability survey-item.

Please note that the US Census Bureau first attempts to have the survey filled via mail correspondence; if that fails, they move to computer-assisted telephone interviewing, and if that fails they then move to computer-assisted personal interviewing (Brault, 2009). In an attempt to reduce individual item non-response, the US Census Bureau also conducts failed-edit follow-up and telephone questionnaire assistance operations (Brault, 2009). The effort made by the US Census Bureau to produce accurate information is admirable. Readers should not be dissuaded from thinking allocation procedures produce unreliable estimates of disability with ACS data. Instead, they should note how even this arguably gold standard data source has limitations that should be kept in mind when producing prevalence rates. The large scale and high quality of data found in ACS makes it tolerable to publicly discuss esoteric nuances of imprecision in the data. In contrast, it is possible that some widely used studies on disability could not stand such public scrutiny without losing face validity.

Because the US Census Bureau is confident about their products, they provide public data users with allocation “flag” variables (variables that can help identify when an allocation is present for a particular response) in PUMS files indicating if a value was observed or allocated (Siordia, 2013b). These procedures are undertaken under the direction of the Federal Committee on Statistical Methodology (FCSM) which has informed researchers that survey accuracy encompasses both sampling error and a broad spectrum of non-sampling-related errors (e.g., item non-response) (OIRA-SPO, 2001). The weighted number of allocations and percent allocated are calculated with the following equation: [(weighted allocated count ÷ total weighted population)×100]. The goal of introducing this last measure is to inform the reader that even though the largest available sample with quasi-pre-clinical disability items for the US is being used in this study, the estimates also have other forms of error which may bias estimates away from true value. This form of bias on the estimates may be statistically unquantifiable (Siordia & Young, 2013).

Despite these serious issues, the disability “population estimates” provided here may be amongst the most reliable—notwithstanding the ambiguity of the survey question. Many other studies, using problematic sample selection designs that severely limit their generalizability, make use of a few thousand people from a small US geographical region to provide what they argue is acceptable ‘population estimates of X-disability.’ The current study uses more than nine million people and provides transparent estimates by framing them through a discourse that includes their SEs, MOEs, RUs, and allocations. Please note throughout the discussion of the findings that only qualitative comparisons between estimates are made. I do not conduct quantitative testing to determine if the differences between groups (e.g., non-Latino-Whites versus Mexican-Latinos) are statistically significant. I deliberately avoid conducting tests to determine “statistically significant differences” to reiterate that disability population estimates have a range of uncertainty.

Findings

I first review the general distribution of the analytic sample presented in Table 1. Both weighted counts (using person-weights) and unweighted counts (using actual number of observations) are presented in Table 1. Instead of reviewing all the numbers, a descriptive discussion on the Person Inflation Ratio (PIR) shown in Table 1 is used. As explained earlier, PIR represents the average number of individuals each of our study participants represents and is calculated by dividing the weighted count by the unweighted number. For example, the PIR for Mexican-Latinos/as is computed as follows: [32,862,818 ÷ 810,696] = 40.54 ≈ 41. This means that on average, each actual Mexican-Latino/a in the microdata represents about 41 other Mexican-Latinos/as after population weights are used to generalize the sample characteristics to the whole of the US population. From the racial-ethnic groups, a qualitative comparison shows MLs and NMLs have the highest PIR at 41, followed by NLB (40), NLO (35), and NLW (31). MLs and NMLs on average represent 47 other MLs and NMLs respectively. From the demographic variables, we see that males; those at or below the age of 49; southern states; those with less than one year of college; and non-US-citizens have the highest PIR values.

Table 1.

Weighted and unweighted estimates from the analytic sample

	Weighted1	Unweighted2	PIR3
Race-Ethnicity
Non-Latino White	196,951,654	6,326,562	31
Non-Latino Black	37,801,756	948,585	40
Non-Latino Other	23,803,937	686,797	35
Mexican-Latino	32,862,818	810,696	41
Non-Mexican Latino	17,811,080	431,797	41
Demographics
Female	157,199,566	4,725,640	33
Male	152,031,679	4,478,797	34
Age ≤ 49	209,708,347	5,753,859	36
Age 50-64	59,026,386	1,955,685	30
Age ≥ 65	40,496,512	1,494,893	27
US-Born	269,337,364	8,174,101	33
Naturalized	17,601,889	502,953	35
Non-Citizen	22,291,992	527,383	42
≤ 1 year of college	219,810,714	6,483,206	34
≥ Associates degree	77,559,393	2,409,301	32
Region by Weather
State in “Warmer South”	117,899,464	3,456,228	34
States in “Colder North”	191,331,781	5,748,209	33

¹Counts using the PWGTP (ACS person-weight) variable

²Counts not using the PWGTP (ACS person-weight) variable

³Person Inflation Ratio (PIR)= (weighted count ÷ weighted total population)

In_PM estimates are presented in Table 2. The number of individuals having difficulties with independent living, by racial-ethnic groups, is as follows: NLWs=5.24%; NLBs=5.67%; NLOs=3.49%; MLs=2.52%; and NMLs=3.66%. As is evident from these population-relative percentages, NLBs have the highest concentration of independent living difficulties and MLs have the least. In absolute terms, NLWs have the largest number (weighted n=10,320,245) of individuals experiencing independent living difficulties. Please note that uncertainty in the In_PM estimates increase in the following order: NLWs; NLBs; MLs; NMLs; and NLOs. In other words, the highest level of imprecision is located in NLOs. Survey item allocations increase in the following order: MLs; NLWs; NLOs; NLMs; and then NLBs. This means the largest number of allocations occurs in the NLB group. Females have more In_PM difficulties than males—their estimate has a smaller RU and they have lower rates of allocation than males. RU is lowest amongst those aged 65 and up (18.50% have an In_PM difficulty)—although their age group has the largest number of allocation rates compared to those at or below age 64. In terms of citizenship status, In_PM estimates for the naturalized group show they have the highest level of In_PM difficulties (5.89%), the largest level of uncertainty (RU=4.01%), and the largest percent of allocation rates (3.83%). Those with one year or less of college education have the most In_PM difficulties (5.73%), lowest RU (0.54%), and largest allocation rate (2.80%), compared to people with an Associates degree or above. Colder North states have greater levels (4.84%) of people experiencing InPM difficulties—although estimates for the warmer South states have higher uncertainty and allocations associated with independent living difficulty measures.

Table 2.

Weighted estimates, their margins of error and allocation rates for difficulty with “independent living” survey item

	Independent Living
	Disable			MO				Allocat
Race-Ethnicity	d1	%D2	SE3	E4	LCL5	UCL6	RU7	ed8	%A9
Non-Latino	10,320,	5.24	26,2	43,1	10,277,	10,363,	0.76	4,590,	2.33
White	245	%	37	60	085	405	%	608	%
	2,144,3	5.67	10,1	16,7	2,127,5	2,161,0	1.42	1,322,	3.50
Non-Latino Black	44	%	83	52	92	96	%	643	%
	829,59	3.49	12,2	20,0	809,52	849,66	4.41	685,74	2.88
Non-Latino Other	7	%	02	72	5	9	%	0	%
	829,09	2.52	7,46	12,2	816,82	841,36	2.70	716,01	2.18
Mexican-Latino	6	%	0	72	4	8	%	0	%
Non-Mexican	652,14	3.66	6,13	10,0	642,04	662,23	2.82	518,13	2.91
Latino	1	%	7	95	6	6	%	1	%
	Disable			MO				Allocat
Demographics	d		SE	E	LCL	UCL	RU	ed	%A
	8,965,2	5.70	18,0	29,7	8,935,5	8,994,9	0.61	3,931,	2.50
Female	44	%	80	42	02	86	%	006	%
	5,810,1	3.82	15,8	26,0	5,784,0	5,836,2	0.82	3,902,	2.57
Male	79	%	61	91	88	70	%	126	%
	3,800,6	1.81	14,5	23,9	3,776,6	3,824,5	1.15	4,546,	2.17
Age < 49	29	%	67	63	66	92	%	873	%
	3,483,9	5.90	11,8	19,5	3,464,3	3,503,4	1.02	1,816,	3.08
Age 50-64	27	%	96	69	58	96	%	562	%
	7,490,8	18.5	14,8	24,4	7,466,4	7,515,2	0.59	1,469,	3.63
Age ≥ 65	67	0%	41	13	54	80	%	697	%
	13,181,	4.89	26,1	42,9	13,138,	13,224,	0.59	6,546,	2.43
US-Born	926	%	16	61	965	887	%	526	%
	1,037,1	5.89	13,8	22,7	1,014,3	1,059,9	4.01	674,05	3.83
Naturalized	80	%	53	88	92	68	%	5	%
	556,31	2.50	5,81	9,56	546,75	565,88	3.14	612,55	2.75
Non-Citizen	7	%	4	4	3	565,881	%	1	%
≤ 1 year of	12,603,	5.73	22,7	37,4	12,566,	12,640,	0.54	6,146,	2.80
college	537	%	45	16	121	953	%	747	%
≥ Associates	2,171,8	2.80	9,84	16,1	2,155,6	2,188,0	1.36	1,686,3	2.17
degree	86	%	2	90	96	76	%	385	%
Region by Weather	Disable			MO				Allocat
	d		SE	E	LCL	UCL	RU	ed	%A
States in “Warmer South”	5,510,5	4.67	15,9	26,4	5,484,1	5,537,0	0.87	3,221,4	2.73
	91	%	55	26	65	17	%	85	%
States in “Colder North”	9,264,8	4.84	18,5	30,4	9,234,3	9,295,2	0.60	4,611,64	2.41
	32	%	04	39	93	71	%	7	%

¹Weighted number of people reporting difficulty with independent living (i.e., “disabled”)

²Percent disabled (%D)= [(weighted disabled count ÷ weighted total population) × 100] (Note: total population available in Table 1)

³Standard Error (SE)

⁴Margin of Error (MOE)

⁵Lower Confidence Limit (LCL): Low limit of 90% confidence interval= [Disable - MOE]

⁶Upper Confidence Limit (UCL): High limit of 90% confidence interval= [Disable + MOE]

⁷Range of Uncertainty (RU) = {[(SE × 3) ÷ disabled] × 100}

⁸Allocated: Number of responses to ‘independent living’ survey item which are assigned or changed

⁹Percent Allocated (%A)= [(weighted allocated count ÷ weighted total population) × 100] (Note: total population available in Table 1)

Out_PM estimates are shown in Table 3. The estimates show that ambulatory difficulty, by racial-ethnic groups, is as follows: NLWs=7.52%; NLBs=8.35%; NLOs=4.42%; MLs=3.73%; and NMLs=5.05%. As with In_PM, NLBs have the highest concentration of ambulatory difficulties and MLs have the least. In absolute terms, NLWs have the largest number (weighted n=14,808,770) of individuals with ambulatory difficulties by virtue of their population size in the US. The uncertainty in the Out_PM estimates increases as follows: NLWs; NLBs; MLs; NLOs; and NMLs. The highest level of imprecision is located in NMLs. Survey item allocations increase as follows: NLWs; MLs; NMLs; NLOs; and NLBs. This means the largest number of allocations is concentrated in NLBs. When compared to males, females have more (7.99%) difficulties with Out_PM than males, although their estimates have a larger RU and slightly lower rate of allocation than males. More than one-fourth (25.78%) of individuals aged 65 and over have difficulties with Out_PM—their RUs are lower than those aged 64 and below, but their age group has the largest number of allocation rates. Non-citizens, with 3.31% having Out_PM difficulties, have the largest level of uncertainty (RU=2.99%), while naturalized individuals have the largest percent of allocation rates (4.04%). On Out_PM, those with one year or less of college education have the highest rate at 8.04%, lowest RU (0.45%), and largest rate of allocation at 3.42%, when compared to those with an Associates degree or more. Warmer South states, with a 6.78% Out_PM rate of difficulty in the population, have greater levels of uncertainty and allocations associated with ambulatory difficulty measures.

Table 3.

Weighted estimates, their margins of error and allocation rates for difficulty with “ambulatory” survey item

	Ambulatory
	Disable			MO				Allocat	%A9
Race-Ethnicity	d1	%D2	SE3	E4	LCL5	UCL6	RU7	ed8
Non-Latino	14,808,	7.52	27,9	45,9	14,762,	14,854,	0.57	5,269,	2.68
White	770	%	55	86	784	756	%	698	%
	3,157,1	8.35	13,8	22,7	3,134,4	3,179,9	1.31	1,538,	4.07
Non-Latino Black	91	%	17	30	61	21	%	750	%
	1,052,7	4.42	7,88	12,9	1,039,7	1,065,7	2.25	838,77	3.52
Non-Latino Other	31	%	4	69	62	00	%	0	%
	1,225,9	3.73	7,46	12,2	1,213,6	1,238,1	1.83	919,59	2.80
Mexican-Latino	15	%	3	77	38	92	%	4	%
Non-Mexican	898,84	5.05	9,29	15,2	883,55	914,12	3.10	613,51	3.44
Latino	0	%	4	89	1	9	%	6	%
	Disable			MO				Allocat
Demographics	d		SE	E	LCL	UCL	RU	ed	%A
	12,566,	7.99	34,0	56,0	12,510,	12,622,	0.81%	4,581,	2.91
Female	126	%	73	51	075	177	%	660	%
	8,577,3	5.64	17,2	28,3	8,548,9	8,605,7	0.60	4,598,	3.02
Male	21	%	59	91	30	12	%	668	%
	4,291,9	2.05	14,9	24,5	4,267,3	4,316,4	1.04		0.03
Age < 49	01	%	31	62	39	63	%	59,724	%
	6,412,9	10.8	23,2	38,3	6,374,6	6,451,2	1.09	1,780,	3.02
Age 50-64	65	6%	99	27	38	92	%	750	%
	10,438,	25.7	23,4	38,5	10,400,	10,477,	0.67	1,427,	3.52
Age ≥ 65	851	8%	35	50	301	401	%	138	%
	19,061,	7.08	33,9	55,8	19,006,	19,117,	0.53	7,809,	2.90
US-Born	922	%	54	54	068	776	%	003	%
	1,342,7	7.63	10,6	17,5	1,325,1	1,360,3	2.39	710,68	4.04
Naturalized	43	%	91	87	56	30	%	2	%
	738,78	3.31	7,36	12,1	726,66	750,89	2.99	660,64	2.96
Non-Citizen	2	%	3	13	9	5	%	3	%
≤ 1 year of	17,671,	8.04	26,7	43,9	17,627,	17,715,	0.45	7,520,	3.42
college	037	%	42	90	047	027	%	697	%
≥ Associates	3,472,4	4.48	23,2	38,2	3,434,1	3,510,6	2.01	1,659,	2.14
degree	10	%	69	78	32	88	%	631	%
	Disable			MO				Allocat
Region by Weather	d		SE	E	LCL	UCL	RU	ed	%A
States in “Warmer South”	7,988,9	6.78	31,6	52,0	7,936,9	8,040,9	1.19	3,782,	3.21
	76	%	13	03	73	79	%	733	%
States in “Colder North”	13,154,	6.88	20,7	32,2	13,122,	13,186,	0.47	5,397,	2.82
	471	%	99	14	257	685	%	595	%

¹Weighted number of people reporting ambulatory difficulties (i.e., “disabled”)

²Percent disabled (%D)= [(weighted disabled count ÷ weighted total population) × 100] (Note: total population available in Table 1)

³Standard Error (SE)

⁴Margin of Error (MOE)

⁵Lower Confidence Limit (LCL): Low limit of 90% confidence interval= [Disable - MOE]

⁶Upper Confidence Limit (UCL): High limit of 90% confidence interval= [Disable + MOE]

⁷Range of Uncertainty (RU) = {[(SE × 3) ÷ disabled] × 100}

⁸Allocated: Number of responses to ‘ambulatory’ survey item which are assigned or changed

⁹Percent Allocated (%A)= [(weighted allocated count ÷ weighted total population) × 100] (Note: total population available in Table 1)

Conclusions

There are some limitations within the current study. For those interested in “health disparities” (Horner-Johnson, Dobbertin, Lee, & Andresen, 2013; Reichard, Stolzle, & Fox, 2011), the between-group comparisons provided in the discussion are only qualitative—I do not claim any of the observed differences are statistically significant. Please note that some do compare estimates by observing whether their confidence intervals overlap to determine if statistically significant differences exist. Statistical significance may not always be relevant in the policy world (Ward, Greenhill, & Bakke, 2010). It may be that an unwarranted amount of attention is paid to finding statistically significant relationships, while too little attention is given to finding practical means of improving the structural systems affecting the disabled. Because stakeholders may want to see high quality estimates, or acceptable estimates whose complexity is reduced to digestible interpretations, the tables in this report may provide such a product: where transparency and easy-to-follow measures highlight proportions of people with difficulties with In_PM and Out_PM in the US population. Because some may be interested in whether the differences are statistically different, future studies should explore this important topic. Future work should also be undertaken to explore these estimates by geographies smaller than the state (e.g., PUMAs).

In general and throughout the qualitative comparisons of In_PM and Out_PM rates, Out_PM is more prevalent than In_PM. It is also found that: NLBs; females; naturalized; colder Northern states; and individuals aged 65 and above have the highest rates of difficulties with both In_PM and Out_PM. In both Out_PM and In_PM, Mexican-Latinos have the lowest concentration of difficulties compared to all the other groups. The range of uncertainty for In_PM estimates is highest amongst NLOs, males, those aged 49 and below, naturalized, and warmer South states. The range of uncertainty for Out_PM estimates is highest amongst NMLs, females, those aged 50-64, non-citizens, and those in warmer South states. For both Out_PM and In_PM, allocation rates are highest amongst NLBs, males, those aged 65 and over, the naturalized, and for those in warmer southern states. Notwithstanding the limitations of the paper, it offers health scientists and policy makers with a previously unavailable source of information on the estimates of disability in the US population. It is my hope that this paper helps with continuing efforts to move society towards a place where inequality, by any compositional or physical characteristic, is mitigated.