Download:
pdf |
pdfPrimary Health Care Patient Surveys
Statistical Design Plan
Table of Contents
1
Introduction...................................................................................................................................... 1
2.
Target Population............................................................................................................................. 2
3.
Grantee Sample Selection ................................................................................................................ 3
3.1 Sampling Frame Construction ................................................................................................. 3
3.2 Stratification Variables ............................................................................................................ 5
3.3 Select Stratified PPS Sample of Grantees ............................................................................... 7
3.4 Grantee Selection Probability ................................................................................................ 11
4.
Site Sample Selection .................................................................................................................... 12
4.1 Determine Eligible Sites within Participating Grantees ........................................................ 12
4.2 Evaluate Distances between Eligible Sites ............................................................................ 12
4.3 Site Selection and Selection Probability................................................................................ 13
5.
Patient Sample Selection................................................................................................................ 14
6.
Sample Sizes and Statistical Power ............................................................................................... 16
7.
Data Collection .............................................................................................................................. 23
7.1 Schedule................................................................................................................................. 23
7.2 Costs....................................................................................................................................... 23
8.
Strengths and Limitations of Study Design ................................................................................... 24
8.1 Strengths ................................................................................................................................ 24
8.2 Limitations ............................................................................................................................. 24
9.
References...................................................................................................................................... 25
i
List of Exhibits
1.
Grantee Characteristics in the Sampling Frame............................................................................... 4
2.
Distribution of Patients Served in 2006 ........................................................................................... 5
3.
Grantee and Patient Yields from Unstratified Random Sampling ................................................... 5
4.
Definition of First-Stage Stratification ............................................................................................ 6
5.
Definition of First-Stage and Second-Stage Stratification............................................................... 6
6.
Definition of Final Stratification...................................................................................................... 7
7.
Grantee Sample Allocation of a Stratified Disproportionate Sampling for the First-Stage
Strata ................................................................................................................................................ 8
8.
Grantee Sample Allocation to Final Strata ...................................................................................... 9
9.
Yield of the Grantee Funding Type and Patients of a Stratified Disproportionate
Sampling ........................................................................................................................................ 10
10.
Grantee and Patient Sample Distribution by Region and Urban/Rural Area................................. 10
11.
Statistical Power to Detect a 10% Difference between the Community Health Population
and the National Health Interview Survey..................................................................................... 17
12.
Statistical Power to Detect a 10% Difference within the Community Health Population ............. 18
13.
Statistical Power to Detect a 10% Difference between the 2002 Community Population
and the 2009 Community Population ............................................................................................ 19
14.
Statistical Power to Detect a 10% Difference between the Community Health Population
and the Public Housing Population................................................................................................ 20
15.
Statistical Power to Detect a 10% Difference between the Migrant Population and the
Homeless Population ..................................................................................................................... 21
16.
Statistical Power to Detect a 10% Difference between the 2002 Homeless Population and
the 2009 Homeless Population....................................................................................................... 22
ii
1. Introduction
The Primary Health Care Patient Surveys (PHCPS), sponsored by the Health Resources and Services
Administration (HRSA), aim to collect data on patients who use health centers funded under Section 330
of the Public Health Service Act. Results from the Patient Surveys will guide and support the Bureau of
Primary Health Care (BPHC) in its mission to improve the health of the Nation’s underserved
communities and vulnerable populations by assuring access to comprehensive, culturally competent,
quality primary health care service. The Patient Surveys will collect data from the clients of health centers
funded through four BPHC grant programs: the Community Health Center Program (CHC), the Migrant
Health Center Program (MHC), the Health Care for the Homeless Program (HCH), and the Public
Housing Primary Care Program (PHPC).
To this end, the BPHC funded two contracts:
•
•
The Community Health Center Patient Survey (CHC) contract will collect and analyze data from
clients of the CHC program. This program serves low-income individuals.
The Health Center Special Populations Patient Survey (Special Populations) contract will collect
and analyze data from clients of the MHC, HCH, and PHPC. Respectively, these three programs
serve migrant workers, homeless individuals, and residents of public housing.
Because some of the Section 330-funded health center grantees (grantees) receive grants through
more than one of the aforementioned grant programs, extensive coordination between the two contracts
will create efficiencies that will allow for larger sample sizes and ensure consistency between the two
studies. Therefore, the sample design for the CHC and Special Populations studies reflects the decision by
BPHC to coordinate these two studies using a harmonized sampling and data collection approach.
In the PHCPS, the primary analytic units are patients who receive services from the funded grantees.
The primary analytic units are clustered within the health center sites (sites) within a grantee. Because
most of the grantees operate more than one site, the sites are clustered within the grantees. RTI
International1 will use a three-stage sample design in which the grantees are selected as the primary
sampling units (PSUs), sites are selected within selected grantees, and patients are selected within
selected sites. Because of the high costs involved with recruiting a grantee and hiring a field interviewer
(FI) to perform the data collection, we will select an independent patient sample from each funding
program for grantees receiving multiple funding programs. The sample design allows us to obtain more
patient interviews with fewer data collection costs due to the high costs of recruiting grantees.
In this report, we summarize the three-staged sample design that will be used for the Patient Surveys.
The sample design will allow for controlled sampling of important characteristics to ensure that certain
comparisons can be made both within the CHC and Special Populations studies, and to other national
studies.
1
RTI International is a trade name of Research Triangle Institute.
1
2. Target Population
The target population for the PHCPS is defined as persons receiving face-to-face services from a
CHC, MHC, HCH, or PHPC grantee, and as persons receiving these services from a clinical staff member
who exercises independent judgment in the provision of services.2 Clients of grantees located within the
50 United States and the District of Columbia are included; clients of grantees within U.S. territories and
possessions are excluded.
Only persons who received services through one of these grantees at least once in the year prior to the
current visit were considered eligible for the survey. This eligibility criterion was used because many of
the questions in the survey ask about services received in the past year; individuals without previous visits
would not have been able to answer these questions and, therefore, were not considered eligible. This
eligibility criterion was also implemented in the BPHC’s previous Community Health Center Survey
(2002) and Healthcare for Homeless Survey (2003).
2
To meet the criterion for “independent judgment,” the provider must be acting on his/her own when serving the
patient and not assisting another provider.
2
3. Grantee Sample Selection
This section discusses the first stage of sample selection, which is the selection of grantees. The
process of selecting grantees includes sample frame construction, stratification, and selection of stratified
probability proportional to size (PPS) grantee samples.
3.1
Sampling Frame Construction
Updated BPHC Uniform Data System (UDS) grantee-level data will be used to construct a sampling
frame for the first stage of selection. The UDS is compiled each year from annual data submissions by
each Section 330-funded grantee. The UDS contains data on key characteristics of the grantees, such as
the type of grant funding received, geographic region, urban/rural location, number of sites within a
grantee, number of patients served, and other information. These same grantee characteristic profiles will
be used in stratification and sample selection. In this statistical design plan, we used data from the 2006
UDS (2007 UDS data have not yet been collected) to illustrate the design plan. Once OMB approval has
been received, the final sample will be drawn using the most current UDS data.
The 2006 UDS data were collected from 1,002 grantees. Some grantees will be excluded from the
sampling frame, including
•
•
•
•
twenty-eight grantees located in U.S. territories or possessions (i.e., those in Puerto Rico, the
Virgin Islands, and the Pacific Basin),
six grantees funded through the CHC program that only operated school-based sites (see
Section 4.1 for more detail on this decision),
seven grantees that received MHC funding only and that served clients through a voucher
program, and
any grantee that is no longer a Section 330-funded grantee.
A total of 961 eligible grantees reporting in 2006 will be included in the grantee sampling frame. We
show the distribution of key grantee characteristics in Exhibits 1 and 2. Exhibit 1 breaks the grantees
down by funding program, region, urban/rural location, and number of sites. In the grantee sampling
frame, 732 grantees had a single funding program, while 229 grantees received funding from multiple
programs. The majority of grantees, roughly 90% and including grantees participating in a single and/or
multiple funding programs, received some CHC funding.
The number of sites within a grantee ranged from 1 to 91. There were 654 grantees that had at least
three sites, with an average of about six sites per grantee. A little over one-third of the grantees had four
to nine sites. The South had 340 grantees, while the West had 265. The Northeast and Midwest had
roughly the same number of grantees with 176 and 180, respectively. Slightly more grantees were in rural
areas than were in urban areas.
Another important grantee characteristic is the number of patients in 2006 (Exhibit 2). Among the
961 eligible grantees in the grantee sampling frame, the number of patients receiving at least one face-toface encounter for services during 2006 varied among the grantees, ranging from 139 to 203,556 and
averaging 15,168. The total number of patients was approximately 14.6 million.
3
Exhibit 1.
Grantee Characteristics in the Sampling Frame
Domain Category
Number of Grantees
Percent Distribution
Funding Program Received
C
H
M
P
CH
CM
CP
PH
CMH
CMP
CPH
CMPH
Total
651
68
9
4
82
98
15
5
16
2
9
2
961
67.74%
7.08%
0.94%
0.42%
8.53%
10.20%
1.56%
0.52%
1.66%
0.21%
0.94%
0.21%
100%
Regiona
Northeast
Midwest
South
West
Total
176
180
340
265
961
18.31%
18.73%
35.38%
27.58%
100%
Urban/Rural Location
Urban
Rural
Total
467
494
961
48.60%
51.40%
100%
Number of Sites
1
2
3
4–9
10–14
15–19
≥20
Total
160
147
125
353
93
42
41
961
16.65%
15.30%
13.01%
36.73%
9.68%
4.37%
4.27%
100%
NOTE: C = Community Health Center Program; H = Healthcare for Homeless Program; M = Migrant
Health Center Program; P = Public Housing Primary Care Program.
a
“Region” refers to Census region here.
4
Exhibit 2.
Distribution of Patients Served in 2006
Patient Distribution
Number of Patients
Range of Number of Patients
Minimum
139
25th Percentile (Q1)
5,1272
Median
10,321
75th Percentile (Q3)
19,539
Maximum
203,556
Mean Number of Patients per Grantee
15,168
Total Number of Patients Across All Grantees
3.2
14,561,166
Stratification Variables
Comparing the CHC survey to the National Health Interview Survey and comparing survey outcomes
between funding programs are the primary analytic goals for BPHC. Therefore, our sample design goals
are as follows:
•
•
Select a sufficient number of patients to complete 4,522 interviews (2,210 for the CHC contract
and 2,312 for the Special Populations contract, as per the modified contract).
Within Special Populations, maintain roughly the same number of patient interviews for both
HCH and MCH with slightly fewer from the PHPC.
As shown in Section 2, the majority of grantees receive CHC funding, while relatively few grantees
receive PHPC and/or MHC funding. A random selection of grantees without any stratification would
result in very small grantee sample sizes, and consequently small patient sample sizes, for the MHC and
PHPC funding programs (Exhibit 3).
Exhibit 3.
Grantee and Patient Yields from Unstratified Random Sampling
Grantee Funding Type
Number of Grantees
Selected
C
105
3,255
H
22
682
M
15
465
P
4
124
146
4,526
Total
Expected Number of Complete
Interviews
NOTE: C = Community Health Center Program; H = Healthcare for Homeless Program; M = Migrant
Health Center Program; P = Public Housing Primary Care Program.
In this scenario, the number of selected grantees is determined using proportional allocation by the
number of grantees for each funding type, as shown in Exhibit 1. A simple random sample of grantees is
selected. The result is displayed in Exhibit 3. In this selection scenario, we select 115 unique grantees. If
a selected grantee participates in multiple funding programs, we would take an independent sample of
each funding program. For example, if a grantee receiving both CHC and MHC funding is recruited, this
grantee would be counted as a CHC grantee and also as an MHC grantee. Therefore, there are 105 CHC
grantees, which count as more than 70% of the total yielded grantees. The PHPC program has only 4
grantees. To calculate the expected completed interviews, we further assume the same number of
completed patient interviews is obtained from each funding program in a grantee and 4,522 total
5
completed interviews for both studies. The patient sample size for MHC and PHPC is very small.
Ultimately, we would have very limited statistical power to perform comparisons.
To facilitate the comparison of survey outcomes between funding programs, a stratified sampling
method with different sampling rates for selecting grantees within each stratum is necessary. To this end,
we will create four mutually exclusive strata by grouping grantees according to the types of funding they
receive. These four groups will serve as the first-stage strata and are defined as follows:
•
•
•
•
Stratum 1: Grantees with CHC Funding Only.
Stratum 2: All Grantees with PHPC Funding.
Stratum 3: Remaining Grantees with MHC Funding.
Stratum 4: All Remaining Grantees Not Included in Strata 1–3.
The number of grantees within each stratum is displayed in Exhibit 4.
Exhibit 4.
Definition of First-Stage Stratification
Grantee Funding Type
Number of
Grantees
Stratum 1: Grantees with CHC Funding Only
C
651
Stratum 2: All Grantees with PHPC Funding
P; CP; PH; CMP; CPH;
CMPH
37
M; CM; CMH
123
H; CH
150
First-Stage Strata
Stratum 3: Remaining Grantees with MHC Funding
Stratum 4: All Remaining Grantees Not Included in Strata 1–3
Total
961
NOTE: C = Community Health Center Program; H = Healthcare for Homeless Program; M = Migrant
Health Center Program; P = Public Housing Primary Care Program.
The above first-stage strata are used to ensure that the selected grantees are representative to the four
funding programs. To ensure the grantees with single funding type of MHC or HCH are represented in the
grantee sample, we split Stratum 3 and Stratum 4 into two second-stage strata as shown in Exhibit 5:
Stratum 3.1, Stratum 3.2, Stratum 4.1, and Stratum 4.2.
Exhibit 5.
Definition of First-Stage and Second-Stage Stratification
Grantee Funding Type
Number of
Grantees
Stratum 1: Grantees with CHC Funding Only
C
651
Stratum 2: All Grantees with PHPC Funding
P; CP; PH; CMP; CPH;
CMPH
37
First-Stage and Second-Stage Strata
Stratum 3: Remaining Grantees with MHC Funding
M; CM; CMH
Stratum 3.1: CM and CMH Grantees
CM; CMH
Stratum 3.2: M Grantees
M
Stratum 4: All Remaining Grantees Not Included in Strata 1–3
Stratum 4.1: CH Grantees
Stratum 4.2: H Grantees
9
H; CH
CH
H
Total
114
82
68
961
NOTE: C = Community Health Center Program; H = Healthcare for Homeless Program; M = Migrant
Health Center Program; P = Public Housing Primary Care Program.
6
Furthermore, to ensure the selected grantee sample within six first-stage and second-stage strata are
representative of grantees with different patient sizes, we further split six strata into several third-stage
strata according to the patient size of a grantee. In each of six strata, we calculate the 33rd and 66th
percentile of patient size. Grantees with patient sizes over the 66th percentile are defined as “Large”
grantees, grantees with patient sizes below the 33rd percentile are defined as “Small” grantees, and
grantees with patient sizes between the 33rd and 66th percentiles are defined as “Medium” grantees. In
order to have the minimum sample size be larger than 10 in each final stratum, some first-, second-, and
third-stage strata are collapsed due to small sample size. Thus, there are a total of 12 final strata in the
grantee sample stratification, as shown in Exhibit 6.
Exhibit 6.
Definition of Final Stratification
Three-Stage Strata
Stratum 1: Grantees with CHC Funding Only
Stratum 1.1.1: Large
Grantee Funding
Type
Final Strata
Number of
Grantees
C
Stratum 1.1.2: Medium
2
Stratum 1.1.3: Small
3
179
234
238
4
37
5
Stratum 2: All Grantees with PHPC Funding
Stratum 2.1.1: Original Stratum 2
Stratum 3: Remaining Grantees with MHC Funding
Stratum 3.1: CM and CMH Grantees
Stratum 3.1.1: Large
1
P; CP; PH; CMP;
CPH; CMPH
M; CM; CMH
CM; CMH
Stratum 3.1.2: Medium
6
Stratum 3.1.3: Small
7
72
31
11
8
9
9
10
51
31
11
12
19
49
961
Stratum 3.2: M Grantees
Stratum 3.2.1: Original Stratum 3.2
Stratum 4: All Remaining Grantees Not Included in
Strata 1-3
Stratum 4.1: CH Grantees
Stratum 4.1.1: Large + Medium
Stratum 4.1.2: Small
Stratum 4.2: H Grantees
Stratum 4.2.1: Large
Stratum 4.2.2: Medium + Small
Total
M
H; CH
CH
H
NOTE: C = Community Health Center Program; H = Healthcare for Homeless Program; M = Migrant
Health Center Program; P = Public Housing Primary Care Program.
In addition to the 12 strata for grantee sample selection discussed above, we will sort the sampling
frame by region (Northeast, Midwest, South, and West), urban/rural location, and number of sites per
grantee within each final stratum when applying Chromy’s (1981) probability minimal replacement
sequential PPS selection procedure. Sorting the sampling frame by these key grantee characteristics and
then applying the PPS sequential procedure will provide implicit stratification according to the order of
the units in a stratum. The selected grantee samples will be distributed among various regions, urban/rural
locations, and number of sites.
3.3
Select Stratified PPS Sample of Grantees
In the multi-stage sample design, when the cluster size varies greatly, the unequal probability
sampling of cluster within each stratum will result in estimates of population characteristics, especially
population totals that have lower variance than those obtained from clusters with equal probability (Levy
7
and Lemeshow, 1999). As mentioned in Section 3.1, the grantees differ widely with respect to the number
of patients served. PPS sampling is a commonly used method of unequal probability sampling in which
the probability of a cluster being sampled is proportional to the level in that cluster of some size measure.
We will use PPS sampling to select the grantee sample from each final stratum. The size measurement
will be the number of patients who visited the grantee for services.
Before selecting a grantee sample from each final stratum, we will determine the grantee sample
allocation for each final stratum. We plan to recruit 115 unique grantees to participate in the studies to
achieve our targeted 4,522 completed patient interviews, 2,210 interviews for CHC and 2,312 interviews
for Special Populations. The grantees with PHPC- and MHC-funded programs will be over-sampled,
while grantees with CHC-funded programs will be under-sampled. We start by determining the grantee
sample allocation to the first-stage strata (4 strata), and then allocate the grantee sample to the sub-stratum
at second- and third-stage stratification. The grantee sample allocation determination steps are ordered as
follows:
Step 1:
Step 2:
Step 3:
Step 4:
Step 5:
Step 6:
Step 7:
Step 8:
Step 9:
Select 31 (85% of 37) grantees receiving PHPC funding in Stratum 2.
Subtract the 31 PHPC grantees in Stratum 2 from the 961 total grantees (n = 930).
Calculate the proportion of remaining grantees in Stratum 3 and Stratum 4. The
proportion of grantees in Stratum 3 is 13.23% (123 / 930), and the proportion of grantees
in Stratum 4 is 16.13% (150 / 930).
Subtract the 31 PHPC grantees in Stratum 2 from the total selected grantee sample of 115
(n = 84).
Determine the number of grantees to be selected for Stratum 3 and Stratum 4 using the
percentages calculated in Step 3; the number of grantees in Stratum 3 is 11, and the
number of grantees in Stratum 4 is 14.
Over-sample Stratum 3 by applying an over-sampling ratio of 2.5 (note that this ratio is
determined to have roughly the same number of interviews for MHC and HCH grantees)
to the proportionate sample of 11, resulting in 28 grantees in Stratum 3.
Retain a proportionate sample for Stratum 4 from Step 5, which is 14.
Allocate to Stratum 1 the remaining grantees that have not been allocated to Stratum 2,
Stratum 3, or Stratum 4 (42 grantees).
Determine the grantee sample size for each sub-stratum proportionally to the number of
grantees in each sub-stratum.
As a result, 42 unique grantees are selected from Stratum 1, 31 unique grantees are selected from
Stratum 2, 28 unique grantees are selected from Stratum 3, and 14 unique grantees are selected from
Stratum 4. Exhibit 7 displays the sample allocation of grantees and the sampling rate for each stratum.
The overall sampling rate is 11.97%. The grantee sample allocation to the final 12 strata is shown in
Exhibit 8.
Exhibit 7.
Grantee Sample Allocation of a Stratified Disproportionate
Sampling for the First-Stage Strata
Total Number
of Grantees
Strata
Selected
Grantees
Sampling
Rate
Stratum 1: Grantees with CHC Funding Only
651
42
6.45%
Stratum 2: All Grantees with PHPC Funding
37
31
83.78%
Stratum 3: Remaining Grantees with MHC Funding
123
28
22.76%
Stratum 4: All Remaining Grantees Not Included in Strata 1–3
150
14
9.33%
Total
961
115
11.97%
NOTE: CHC = Community Health Center Program; MHC = Migrant Health Center Program; PHPC =
Public Housing Primary Care Program.
8
Exhibit 8.
Grantee Sample Allocation to Final Strata
Final
Strata
Number of
Grantees
Selected in the
Sample
Stratum 1.1.1: Large
1
12
Stratum 1.1.2: Medium
2
15
Stratum 1.1.3: Small
3
15
4
31
Stratum 3.1.1: Large
5
16
Stratum 3.1.2: Medium
6
7
Stratum 3.1.3: Small
7
3
8
2
Grantee Funding
Type
Three-Stage Strata
Stratum 1: Grantees with CHC Funding Only
C
P; CP; PH; CMP;
CPH; CMPH
Stratum 2: All Grantees with PHPC Funding
Stratum 2.1.1: Original Stratum 2
Stratum 3: Remaining Grantees with MHC Funding
M; CM; CMH
Stratum 3.1: CM and CMH Grantees
CM; CMH
Stratum 3.2: M Grantees
M
Stratum 3.2.1: Original Stratum 3.2
Stratum 4: All Remaining Grantees Not Included in
Strata 1–3
H; CH
Stratum 4.1: CH Grantees
CH
Stratum 4.1.1: Large + Medium
9
5
10
3
Stratum 4.2.1: Large
11
2
Stratum 4.2.2: Medium + Small
12
5
Stratum 4.1.2: Small
Stratum 4.2: H Grantees
H
Total
116
NOTE: C = Community Health Center Program; H = Healthcare for Homeless Program; M = Migrant
Health Center Program; P = Public Housing Primary Care Program.
To account for selected grantees’ refusals to participate, we will select 15% more grantees for each
stratum. We assume an 85% grantee recruitment rate. The remaining 15% of grantees selected will be
held in reserve to replace grantees refusing to participate in the study.
After a PPS selection of grantees in each of the 12 strata is completed, 116 grantees will be in the
sample. As stated in Section 3.2, an independent sample will be selected for each funding program if a
selected grantee has multiple funding programs, which will yield 192 funding programs out of 116
grantees. To achieve interview targets of 2,210 CHC patients and 2,312 special population patients, the
number of complete interviews for each funding type is calculated and displayed in Exhibit 9.3 For this
grantee sample selection scenario, there are 99 CHC grantees, 31 HCH grantees, 31 MHC grantees, and
31 PHPC grantees from which we will select the next stage sample and sites. We discuss this selection
further in Section 4.
3
Note that during the sampling plan implementation, the sample realization may yield a slightly different
distribution of grantees for each funding type.
9
Exhibit 9.
Yield of the Grantee Funding Type and Patients of a Stratified
Disproportionate Sampling
Number of Grantees
for Each Funding
Program
Funding Program
C
Number of Patients per
Grantee for Each Funding
Program
99
Number of Completed
Interviews for Each
Funding Program
23
2,277
H
31
27
837
M
31
27
837
P
31
22
682
Total
192
4,633
NOTE: C = Community Health Center Program; H = Healthcare for Homeless Program; M = Migrant
Health Center Program; P = Public Housing Primary Care Program.
We expect 2,277 interviews for CHC and 2,356 interviews for Special Populations, which is roughly
on target with the goal of 2,210 interviews for CHC and 2,312 for Special Populations (Exhibit 9). For
the Special Populations survey, the number of interviews for PHPC-funded programs is smaller than the
number of interviews for HCH- and MCH-funded programs. The reason for this difference is that the
PHPC patient population is much smaller than the HCH and MCH patient populations. We discuss this
difference in detail in Section 5, where we discuss the patient sample selection.
In Exhibit 10, we display the grantee and patient sample distribution by region and urban/rural area
from the sampling realization discussed above. The distributions of grantee and patient sample by region
and the patient sample distribution by urban/rural area are very similar to the distributions of grantee
frame and patient population. The grantee sample has a slightly higher proportion of urban grantees than
the proportion in the grantee frame. The reason for this difference is that we selected 31 PHPC grantees
out of 37, and those PHPC grantees are mainly in urban areas (35 in urban areas, 2 in rural areas).
Therefore, our proposed grantee sample selection and patient sample selection methods produced grantee
and patient samples that represented the target population in different regions and urban/rural areas very
well.
Exhibit 10. Grantee and Patient Sample Distribution by Region and
Urban/Rural Area
Grantee Frame
Domains
Grantee Sample
Patient Population
%
n
%
N
961
100%
116
100%
14,561,166
100%
4,633
100%
Northeast
176
18.3%
23
19.8%
2,942,832
20.2%
931
20.1%
Midwest
180
18.7%
22
19.0%
2,672,756
18.4%
839
18.1%
South
340
35.4%
37
31.9%
4,516,264
31.0%
1,521
32.8%
West
265
27.6%
34
29.3%
4,429,314
30.4%
1,342
29.0%
961
100%
116
100%
14,561,166
100%
4,633
100%
Urban
467
48.6%
65
56.0%
8,236,600
56.6%
2,643
57.0%
Rural
494
51.4%
51
44.0%
6,324,566
43.4%
1,990
43.0%
Region
Urban/rural
%
Patient Sample
N
n
%
When we select the real sample for the study, to meet the targeted number of complete interviews of
4,522 (2,210 for CHC and 2,312 for Special Populations), we may need to adjust the sampling rates at
grantee selection for each stratum described previously and the number of interviews per grantee for a
specific funding program.
10
3.4
Grantee Selection Probability
The selection probability for the ith grantee within the hth stratum is given by
G
hi
=n
S
hi ,
h ∑S
hi
i
(1)
where h is the index for the strata (Stratum 1, Stratum 2, Stratum 3, and Stratum 4); i is the index for
grantees on the frame within each stratum; nh is the number of grantees to select in the hth stratum; and Shi
is the size measure, which is the number of patients in each grantee.
We are aware that applying different sampling rates for each stratum and taking grantee samples
through PPS sampling causes an increase of variability of the selection probability. Consequently, this
application of different sampling rates increases variability of sampling weights, namely increases the
unequal weighting effect (UWE) on the variances of sample estimates and reduces the statistical power of
the analysis. To minimize the impact of UWE, we will select sites within grantees using PPS sampling in
the second stage of selection, and we will select the same number of patients per grantee in the third stage
of sample selection. We address these issues in more detail in Section 4 and Section 5.
11
4. Site Sample Selection
As discussed previously, more than half of the grantees have three or more sites, and, in general,
those grantees with more sites tend to have more patients. Furthermore, the grantees are selected with the
PPS method at the first stage of selection, which means that grantees with large numbers of patients have
a higher probability of being selected in the sample. As a result, we expect a fair number of the grantees
recruited to have more than three sites. We will allow at most three sites for each funding program within
a grantee to be in the PHCPS; therefore, for those grantees with more than three sites, we will select three
from their larger total. This section discusses the second stage of selection, which is the selection of sites
from participating grantees that have multiple sites.
4.1
Determine Eligible Sites within Participating Grantees
Once a grantee is recruited and agrees to conduct the study in its sites, our recruiters will work with
the grantee’s administration to identify eligible sites. The following eligibility criteria will be used:
•
•
•
•
The site should participate in at least one of the four specific funding programs and must have
been operating under the grantee for at least 1 year.
The site is not a temporary clinic.
The site is not a school-based health center.
The site is not a specialized clinic, excepting clinics providing OB/GYN services.
Due to the complexity of recruiting school-based sites, including the extensive efforts associated with
getting permission from schools and parents/guardians to interview the adolescent patients, recruiting
stand-alone, school-based sites is not feasible within the current survey schedule and budget. Therefore,
such sites will be excluded from the Patient Surveys. Although these sites are excluded, we will not
necessarily be excluding all patients who receive school-based health services. Some children who
receive medical care at a school-based health center site may also receive some of their care at a nonschool-based Community Health Center site near their residence.
After the eligible sites are identified, we will collect or verify the following information from/with
each participating grantee:
•
•
•
4.2
number of eligible sites serving each client type (i.e., migrants, homeless, public-housing, and
low-income);
address and contact information for each eligible site; and
number of patients served in each eligible site, overall and by type of client.
Evaluate Distances between Eligible Sites
In most cases, only one field interviewer (FI) will be hired to collect data within each participating
grantee. Therefore, it is desired that sites are within manageable distances for the FI(s). The grantees tend
to operate sites in relatively localized areas. Experience from the 2002 Community Health Center Survey
showed that, out of the 70 grantees selected for the study, 82% of the sites selected were within 30 miles
of the grantee’s headquarters, with an additional 9.2% of the sites being within 45 miles of the central
health center location. The 2002 Community Health Center data suggest that the majority of the selected
sites are within an area that can be covered by one FI without incurring extensive travel costs. We expect
that the same will be true for the current surveys. However, our sampling staff will evaluate distances
between the administrative office/central site and the associated sites as soon as possible to determine if
any modifications are needed to the selection of sites within the grantee, or if special data collection
arrangements should be made. We will pay special attention to sites that are located more than 100 miles
from the administrative office/central site. The Project Officer will be consulted if issues of distance arise.
12
4.3
Site Selection and Selection Probability
If there are three or fewer sites for a population type (i.e., migrant, homeless, public-housing, and
low-income) and all of the sites are within a manageable distance by one FI, all of the sites will be
included in the study. If one site is far from the other sites and the other sites are close to each other, the
two sites that are close to each other will be selected. However, if all three sites are far from each other,
we will select the site that has the largest patient volume. Similarly, when two sites for a specific funded
program are far from each other, the one with the largest number of patients will be selected. Again, these
special cases will be reviewed with the Project Officer.
For grantees with more than three sites for a population type, we will use a PPS sampling method
similar to the one discussed in Section 3.3 to select three sites from the sites within a manageable distance
for one FI to cover. The number of patients for the sites under a specific funding program will serve as the
size measure in the PPS sampling. For the grantees who participate in multiple funding programs, an
independent PPS selection of sites will be conducted for each funding program, if needed.
The selection probability for the jth site within the ith grantee for f funding program is given by
C fij
⎧
⎪
⎪ 1 , if 3 or fewer sites are all selected, or
⎪
=⎨
⎪ 3s fij
, if 3 sites are selected through PPS sampling,
⎪
⎪ ∑ s fij
⎩ j
where sfij is the number of patients in site j within grantee i for funding program f.
13
(2)
5. Patient Sample Selection
Because of the mobile nature of some of the target populations of this study, a random sample of
patients will be chosen for interviews as they enter the site and register with the receptionist for services.
A field interviewer will visit a selected site for a predetermined number of days in the sampling period to
conduct interviews. The receptionist will be instructed to select the first eligible patient registered after
the FI informs the receptionist that he/she is ready for the next interview. The receptionist will read to the
selected patient a brief script about the study and direct the patient to the FI for questions or participation.
The receptionist will be asked to keep track of the number of patients who enter the site and the number
of patients selected while the FI is at the site to conduct data collection. The receptionist can either use
tally marks to count patients as they enter or complete a table based on the sign-in sheet or appointment
list before the FI leaves the site. The patient count sheets for each FI data collection visit will be sent to
RTI for data entry, and counts will be used to calculate the analysis weights for the study.
To minimize the UWEs of selecting a PPS sample, the same number of patients will be selected from
the grantees in each of the funding programs. As shown in Exhibit 9 in Section 3.3, 99 CHC grantees, 31
MHC grantees, 31 HCH grantees, and 31 PHPC grantees are recruited for CHC and Special Populations.
To achieve 2,210 completed interviews, we expect 23 patient interviews will be completed from each
participating CHC grantee. For Special Populations, we will achieve 2,312 completed interviews.
However, the goal of roughly the same interviews for MCH, HCH, and PHPC is difficult to achieve. In
the 2006 UDS, there were 701,623 patients from the HCH program and only 129,280 patients from the
PHPC program. The PHPC population is much smaller than the HCH population. We are concerned
about the amount of time an interviewer may need to spend in the sites in order to contact eligible PHPC
patients. Therefore, we have reduced the production goal for the PHPC population. We expect that 27
patient interviews will be completed from each participating HCH and MHC grantee, and 22 patient
interviews will be completed for each participating PHPC grantee. The reduced goal of 22 completed
interviews per PHPC grantee may still be too aggressive. We will check the patient volume for each
selected site for served PHPC patients and consult with the BPHC Project Officer if this goal becomes
problematic.
Within each grantee, if more than one site is selected into the study for a specific funding program,
the number of patient interviews within that grantee will be divided equally among those sites. For
example, if three sites are selected within an HCH grantee, 16 patients will be surveyed from each site.
If a grantee participates in more than one funding program, independent patient samples will be
selected for each funding program. If a site is chosen for multiple funding programs, the receptionist at
the site will be asked to track and to select patients on the FI visiting dates for all funding programs. The
FI will screen participating patients to determine patient population types (i.e., homeless, migrant,
public-housing, or low-income) and will use the appropriate questionnaire to conduct the patient
interviews.
The selection probability of patient k from within grantee i, site j for funding program f is given by
Pfijk =
n fij
e fij s fij
,
(3)
where nfij is the number of completed interviews from grantee i, site j for funding program f; efij is
observed patient eligibility rate in grantee i, site j for funding program f; and sfij is the number of patients
in grantee i, site j for funding program f.
14
The probability of a patient selected in the study is the product of Ghi, Cfij, and Pfjik in Formulas (1),
(2), and (3), respectively. That is
π hfijk =
nh s hi 3s fij n fij
.
∑ s hi ∑ s fij e fij s fij
i
(4)
j
For a specific funding program in a grantee that has three sites selected through PPS sampling, the
patient selection probability is maintained in roughly the same manner (self-weighting) within each
stratum, as shown in Formula (4). Because nfij is the same for each site, sfij is cancelled out, and, within a
grantee, the proportion of
s fij and shi is a constant. However, for a specific funding program in a
∑
j
grantee that has three or fewer sites that are all selected, the Cfij is 1. To maintain the same equal selection
probability property, the number of interviews per site (nfij) should be allocated proportionally to the
number of patients of the site rather than equally allocating the interviews to the sites. That is
n fij = n fi
s fij
∑s
,
fij
j
where nfi is the number of interviews from a grantee for funding program f.
15
6. Sample Sizes and Statistical Power
Statistical tests attempt to use data from samples to determine whether a difference exists in a
population or between two populations. An example of a statistical test would be to test the null
hypothesis that the number of uninsured children aged 12 or younger does not differ between the CHC
and MCH populations. The power of the test is the probability that the test will find a statistically
significant difference between two populations as a function of the size of the true difference between
those two populations. There is always a chance that the samples will appear to support or to refute a
tested hypothesis when the reality is the opposite. That risk is quantified as the statistical significance
level. We use significance level of 0.05 to calculate statistical power in this document.
We are using a three-stage sample design in which the grantees are selected as the PSUs, sites are
selected within grantees, and patients are selected within sites. For a specific funding program, up to three
sites within a grantee will be selected. Sample sizes for grantees, sites, and patients are based on an
integrated sample design across the four funding programs. The sample design considers the overlap of
funding programs between sites and the clustering effect of the sites on the demographics of the patients.
The clustering effect and the previously mentioned UWE together contribute to the design effect (deff),
which is a measure of the precision gained or lost by the use of the more complex design instead of a
simple random sample. The design effect is a function of the clustering effect and the UWEs. A design
with a large deff will reduce the statistical power of the analysis.
Results from the 2002 Healthcare for Homeless and Community Health Center User Survey analyses
provide valuable insights into the amount of clustering and deff that might be incurred in the upcoming
CHC and Special Populations studies and into our ability to make comparative analyses to the other study
populations. Selecting patients within sites within grantees produces a clustering effect. We approximated
the intra-cluster correlation (ICC) and the UWE to more accurately gauge the effective sample size and
statistical power between funding programs for the key outcome measures by using the 2002 CHC Survey
data. We used the sample size from the stratified disproportionate sampling scenario from Exhibit 9 and
estimated the sample size for each funding type using demographic information and some key measures
from the 2006 UDS. Exhibits 11–16 display the statistical power of detecting a 10% difference between
surveys of interest for five outcomes. Exhibit 11 displays the statistical power of detecting a 10%
difference between the CHC population and the population in National Health Interview Survey (NHIS).
Exhibit 12 displays the statistical power of detecting a 10% difference between the interested domains
within the CHC population. Exhibit 13 displays the statistical power of detecting a 10% difference
between the CHC population and the previous 2002 CHC population. Exhibit 14 displays the statistical
power of detecting a 10% difference between the CHC population and the PHPC population. Exhibit 15
displays the statistical power of detecting a 10% difference between the MHC population and the HCH
population. Exhibit 16 displays the statistical power of detecting a 10% difference between the HCH
population and the previous 2002 HCH population.
For the comparative analyses between the CHC and other national surveys (such as NHIS), there
should be sufficient power to provide meaningful comparison for all five outcome measures due to the
increase in the sample size of the CHC population and the large sample size from the NHIS as shown in
Exhibit 11. The statistical powers of detecting a 10% difference between the domains within the CHC
population, and comparisons between the CHC population and the previous 2002 CHC population, are
reasonably high for most of the comparisons (Exhibit 12 and Exhibit 13). The statistical power of
detecting a 10% difference between the CHC population and the PHPC population for five outcome
measures is pretty good. When comparing patients diagnosed with hypertension, the chance of detecting
10% of the difference is 83.7%. This percentage indicates sufficient power to detect differences between
members of these populations. When looking at females with hypertension in the Community and Public
Housing populations, the power is reduced to 64.9% due to the reduction in sample size as shown in
Exhibit 14. The ability to detect meaningful differences between the MHC population and the HCH
population and between the HCH population and the previous 2002 HCH population will be somewhat
limited to domains with higher sample sizes as shown in Exhibit 15 and Exhibit 16.
16
Exhibit 11. Statistical Power to Detect a 10% Difference between the
Community Health Population and the National Health Interview
Survey
CHC versus NHIS
Domain
Smoking
Drinking
Asthma
Hypertension
Diabetes
100.0%
100.0%
100.0%
100.0%
100.0%
under 18
95.3%
92.7%
100.0%
96.0%
100.0%
18–34
89.3%
85.2%
99.6%
90.5%
100.0%
35–49
80.4%
75.2%
98.3%
81.9%
99.9%
50+
82.5%
77.5%
98.7%
84.0%
99.9%
NH-White
96.7%
94.6%
100.0%
97.2%
100.0%
NH-Black
85.5%
80.7%
99.2%
86.8%
100.0%
HISP
96.6%
94.5%
100.0%
97.2%
100.0%
Male
98.0%
96.5%
100.0%
98.4%
100.0%
Female
99.8%
99.5%
100.0%
99.9%
100.0%
Insured
99.8%
99.6%
100.0%
99.9%
100.0%
Uninsured
97.8%
96.2%
100.0%
98.2%
100.0%
100.0%
99.9%
100.0%
100.0%
100.0%
92.4%
88.8%
99.8%
93.3%
100.0%
Total
Age Group
Race/Ethnicity
Gender
Insurance Status
Language
English
Non-English
NOTE: Power calculations are based on two-sample t-tests comparing prevalence rates of five outcomes
with a 0.05 level of significance.
17
Exhibit 12. Statistical Power to Detect a 10% Difference within the
Community Health Population
Comparisons within CHC
Domain
Smoking
Drinking
Asthma
Hypertension
Diabetes
Age Group
Under 18 vs. 18–34
Under 18 vs. 35–49
Under 18 vs. 50+
18–34 vs. 35–49
18–34 vs. 50+
35–49 vs. 50+
66.2%
59.2%
60.8%
55.3%
56.6%
51.9%
62.8%
56.1%
57.6%
52.0%
53.3%
48.5%
86.7%
80.0%
81.5%
77.0%
78.6%
74.8%
67.4%
60.4%
61.9%
56.4%
57.8%
53.0%
93.2%
87.8%
89.1%
85.8%
87.2%
84.5%
Race/Ethnicity
NH-White vs. NH-Black
NH-White vs. HISP
NH-Black vs. HISP
64.4%
76.0%
65.6%
61.2%
72.4%
61.5%
84.6%
93.6%
88.4%
65.6%
77.2%
67.0%
91.5%
97.6%
95.1%
Gender
Male vs. Female
87.2%
84.0%
98.5%
88.2%
99.7%
Insurance Status
Insured vs. Uninsured
86.2%
83.5%
97.4%
87.0%
99.2%
Language
English vs. Non-English
80.0%
77.3%
94.1%
80.9%
97.4%
NOTE: Power calculations are based on two-sample t-tests comparing prevalence rates of five outcomes
with a 0.05 level of significance.
18
Exhibit 13. Statistical Power to Detect a 10% Difference between the 2002
Community Population and the 2009 Community Population
CHC09 versus CHC02
Domain
Smoking
Drinking
Asthma
99.2%
98.8%
100.0%
99.4%
100.0%
71.6%
60.6%
49.9%
52.1%
68.0%
57.0%
46.6%
48.8%
91.0%
82.8%
72.2%
74.6%
72.9%
61.9%
51.0%
53.2%
96.2%
90.7%
82.1%
84.2%
Race/Ethnicity
NH-White
NH-Black
HISP
75.3%
55.5%
75.0%
71.7%
52.1%
71.4%
93.1%
78.1%
93.0%
76.5%
56.7%
76.2%
97.3%
87.1%
97.3%
Gender
Male
Female
79.9%
92.2%
76.5%
90.0%
95.4%
99.3%
81.0%
92.9%
98.5%
99.9%
Insurance Status
Insured
Uninsured
92.6%
79.1%
90.4%
75.6%
99.3%
95.0%
93.2%
80.2%
99.9%
98.3%
Language
English
Non-English
95.9%
65.5%
94.4%
61.8%
99.8%
86.7%
96.4%
66.7%
100.0%
93.5%
Total
Age Group
Under 18
18–34
35–49
50+
Hypertension
Diabetes
NOTE: Power calculations are based on two-sample t-tests comparing prevalence rates of five outcomes
with a 0.05 level of significance.
19
Exhibit 14. Statistical Power to Detect a 10% Difference between the
Community Health Population and the Public Housing Population
CHC versus PHPC
Domain
Smoking
Drinking
Asthma
Hypertension
Diabetes
82.8%
80.6%
95.1%
83.7%
97.9%
Under 18
43.3%
41.2%
61.0%
44.2%
69.7%
18–34
33.2%
31.5%
47.7%
33.8%
55.5%
35–49
23.6%
22.5%
33.6%
24.0%
39.3%
50+
21.2%
20.4%
29.8%
21.6%
34.8%
NH-White
16.2%
15.7%
21.8%
16.5%
25.0%
NH-Black
42.6%
40.1%
61.6%
43.5%
71.0%
HISP
45.4%
43.1%
63.4%
46.2%
72.0%
Male
41.8%
39.9%
58.5%
42.6%
66.7%
Female
63.9%
61.3%
82.4%
64.9%
89.0%
Insured
61.0%
58.4%
79.5%
62.0%
86.6%
Uninsured
45.8%
43.7%
63.7%
46.7%
72.1%
English
60.9%
58.4%
79.0%
61.8%
86.0%
Non-English
44.6%
42.2%
63.3%
45.5%
72.2%
Total
Age Group
Race/Ethnicity
Gender
Insurance Status
Language
NOTE: Power calculations are based on two-sample t-tests comparing prevalence rates of five outcomes
with a 0.05 level of significance.
20
Exhibit 15. Statistical Power to Detect a 10% Difference between the
Migrant Population and the Homeless Population
MHC versus HCH
Domain
Smoking
Drinking
Asthma
Hypertension
Diabetes
68.3%
64.5%
89.1%
69.5%
95.1%
Under 18
20.4%
18.8%
34.1%
21.0%
43.8%
18–34
26.0%
24.2%
41.0%
26.6%
50.4%
35–49
22.6%
21.3%
33.5%
23.0%
40.1%
50+
16.2%
15.3%
23.4%
16.5%
28.0%
NH-White
9.9%
9.6%
12.3%
10.0%
13.8%
NH-Black
7.3%
7.2%
8.4%
7.4%
9.0%
32.0%
29.1%
54.9%
33.0%
69.2%
Male
41.1%
38.4%
61.3%
42.0%
71.6%
Female
39.2%
36.3%
60.8%
40.2%
72.2%
Insured
31.0%
28.6%
49.7%
31.8%
61.0%
Uninsured
48.0%
45.0%
69.7%
49.1%
79.6%
English
27.2%
25.9%
38.7%
27.7%
45.2%
Non-English
23.3%
21.3%
41.1%
24.1%
54.1%
Total
Age Group
Race/Ethnicity
HISP
Gender
Insurance Status
Language
NOTE: Power calculations are based on two-sample t-tests comparing prevalence rates of five outcomes
with a 0.05 level of significance.
21
Exhibit 16. Statistical Power to Detect a 10% Difference between the 2002
Homeless Population and the 2009 Homeless Population
HCH09 versus HCH02
Domain
Smoking
Drinking
Asthma
Hypertension
Diabetes
71.4%
67.5%
91.5%
72.7%
96.6%
Under 18
16.2%
15.2%
24.6%
16.5%
30.4%
18–34
26.4%
24.5%
41.7%
27.0%
51.3%
35–49
32.6%
30.3%
51.3%
33.5%
62.1%
50+
22.0%
20.5%
34.6%
22.5%
42.8%
NH-White
32.6%
30.3%
51.3%
33.5%
62.1%
NH-Black
34.1%
31.6%
53.4%
35.0%
64.3%
HISP
22.6%
21.0%
35.5%
23.1%
43.9%
Male
48.1%
44.7%
71.3%
49.2%
81.8%
Female
37.7%
35.0%
58.4%
38.7%
69.6%
Insured
28.1%
26.1%
44.5%
28.8%
54.4%
Uninsured
56.1%
52.5%
79.7%
57.4%
88.8%
English
64.4%
60.5%
86.7%
65.7%
93.8%
Non-English
16.3%
15.3%
24.9%
16.6%
30.8%
Total
Age Group
Race/Ethnicity
Gender
Insurance Status
Language
NOTE: Power calculations are based on two-sample t-tests comparing prevalence rates of five outcomes
with a 0.05 level of significance.
22
7. Data Collection
7.1
Schedule
PHCPS survey data will be collected over a period of 4 months. Although data collection was
originally scheduled for March through May 2009, we will revise our timeline and request a no-cost
extension, per BHPC’s request to allow more time to review the draft questionnaire and receive OMB
clearance. A revised schedule will be provided in the final version of Deliverable 4: Implementation Plan.
To reduce site burden, we will minimize the data collection period at each site. Because the estimated
time to complete each interview with Special Population respondents is 3.5 hours and the estimated time
to complete each interview with CHC respondents is 4.92 (see Section 7.2 for how this estimate was
created), and because the typical interviewer is only available for part-time employment, an average of 1
week of data collection will be required for every six interviews. Production goals, including all four
patient populations, per site, will range from 8 to 99, which means that the field period in any given site
could be as short as 2 weeks or as long as 11 weeks. However, in cases in which the production goal
exceeds 50 patients, and as deemed appropriate by the site, special accommodations can be made, such as
staffing an interviewer who can be at the site full time or bringing in an additional interviewer to help
complete interviews in a shorter period of time.
7.2
Costs
The three primary field costs associated with all completed cases are interviewer labor, mileage
incurred by interviewers, and incentives paid to respondents. Our statistical design and data collection
plans assume interviews will be completed at a rate of 3.5 hours each for Special Population respondents
and 4.92 hours each for CHC respondents. These figures include time for driving to and from a facility,
waiting to be approached by eligible patients, screening potential participants, administering informed
consent, administering an interview, updating field status codes and completing other administrative
paper work, shipping material back to RTI, and participating in regular conference calls with his/her field
supervisor. We assume that interviewers will require reimbursement for an average of 36 miles per
completed interview. Finally, we have budgeted for $25 in incentives for each survey participant.
23
8. Strengths and Limitations of Study Design
Sample designs that maximize the ability to make inferences about a target population will also have
limitations due to budget and schedule constraints. This section addresses the strengths and limitations in
the sample design for the Patient Surveys.
8.1
Strengths
The three-stage PPS sample design will produce a sample of grantees, health centers, and patients that
will spread the samples across the United States and across urban/rural locations and various grantee sizes
according to the numbers of patients and sites. The resulting sample of patients will provide BPHC with
data that will allow them to make references to the patient population receiving services through CHC,
MHC, PHPC, and HCH.
The sample design has stratified the grantees into groups by funding program in order to provide
samples of close to 2,210 patients in CHC and 2,312 patients in Special Populations, as discussed in
Section 5. The sample sizes are based on the selection of 115 unique grantees, but, because we are
allowing patient samples to be selected from each funding program in which the grantee participates, the
patient sample sizes will be equivalent to selecting samples from 192 grantees. This sample design, which
takes advantage of the multiple funding programs received by some of the grantees, results in a 67%
increase of efficiency in recruiting the grantees for the PHCPS. The patient interviews from more grantees
will result in better statistical power than if the patient samples had only been selected from a total of 115
grantees, with each grantee only representing one funding program.
The combined sample of patients from the four funding programs will be sufficient for comparative
analyses with national estimates of U.S. residents from the NHIS on a number of outcomes and
subpopulations of patients and U.S. residents. Comparative analyses between the funding programs may
be adequate for larger subgroups of the patient populations.
8.2
Limitations
Although the sample design takes advantage of the multiple funding programs received by some of
the grantees, due to budget restrictions, the patient sample sizes for each funding program will not be
large enough for comparative analyses of patient characteristics between the funding programs for certain
less-prevalent subgroups of the patient populations. The low statistical power estimates from such
subgroups, such as comparisons of smoking, drinking, asthma, hypertension, and diabetes prevalence
between MHC and HCH, are illustrated in the statistical power exhibits (Exhibits 11–16).
An additional limitation pertains to capturing seasonal variation in health care needs and service
utilization. The time constraints for completing the study within the contract time period will limit the
data collection period to 3 months. Because the data collection period will occur over a period of time that
is less than 1 year, the study will not be able to address any seasonal fluctuations in the types of services
provided to the health center patients during different seasons of the year. The spring data collection will
underestimate patients who enter the health centers for flu shots, typically during late fall, and for coldand flu-related illnesses that typically occur during the winter months. The short time period for data
collection may also miss groups of migrant workers who are migrating to certain areas of the United
States to work in fields that produce crops that need to be harvested in the spring, while some of the
health centers in the study may be in areas where the crops are harvested in the fall.
Finally, for those funding programs in a grantee that has three or fewer sites that are all selected, to
reduce the UWE, the number of patients selected from each site should be proportional to its number of
patients (as discussed in Section 5). For the ease of field operation, we will allocate the patient sample
equally to the sites in a similar way to how the grantees with three sites are selected through PPS
sampling. In doing so, however, we could inflate the UWEs and consequently lose some statistical
analysis power.
24
9. References
Chromy, J. R. (1981). “Variance Estimations for a Sequential Sample Selection Procedure” in
D. Krewski, R. Platek, and J.N.K. Rao, eds. Current Topics in Survey Sampling. New York:
Academic Press, Inc.
Levy, P. S., and Lemeshow, S. (1999). Sampling of Population: Methods and Applications. Third edition.
New York: John Wiley and Sons.
25
| File Type | application/pdf |
| File Title | Microsoft Word - Statistical Design Plan.doc |
| Author | acash |
| File Modified | 2009-03-03 |
| File Created | 2009-03-03 |