Oklahoma Department of Human Services, DAB No. 1436 (1993)

  Department of Health and Human Services

        DEPARTMENTAL APPEALS BOARD

     Appellate Division

SUBJECT:        Oklahoma Department             of   DATE:  August 30,
1993 Human Services Docket No. A-93-59 Decision No. 1436

   DECISION

The Oklahoma Department of Human Services (Oklahoma) appealed an amended
disallowance by the Administration for Children and Families (ACF) of
$612,562 in federal financial participation (FFP) claimed under title
IV-A of the Social Security Act (Act).  The disallowed claim represents
payments made by Oklahoma for child care provided on behalf of families
eligible for Aid to Families with Dependent Children (AFDC) from July
1989 through September 1990. 1/

In its appeal of the amended disallowance, Oklahoma took issue with the
statistical methodology employed by ACF in calculating the disallowance
amount, as well as with ACF's determinations on six of the individual
sample payments.

As discussed below, we find that the statistical methodology employed by
ACF to calculate the disallowance, as revised during the course of this
appeal, was proper.  However, we reverse ACF's disallowance with respect
to four of the six disputed sample payments.  ACF should therefore
recalculate the disallowance in light of our determination that these
four cases were correctly paid, and notify Oklahoma of the revised
disallowance amount.  If Oklahoma disagrees with ACF's recalculations,
it may return to the Board for review of this limited issue within 30
days of receipt of ACF's determination.

Applicable law, regulations, and policy

The Family Support Act of 1988, Public Law No. 100-485, created the AFDC
child care program at section 402(g) of the Act.  Section 402(g)
provides that states must guarantee child care to the extent that the
state determines nececessary for an individual in a family with a
dependent child to (1) accept or maintain employment or (2) participate
in an education and training activity if the state approves the activity
and periodically determines that the individual is satisfactorily
participating in the activity.  The Act provides for states to receive
FFP in their allowable child care expenditures.  Sections 402(g)(3),
403(a) of the Act.

The implementing regulation requires states to guarantee child care for
qualifying children "to the extent necessary to permit an AFDC-eligible
family member" to accept employment or remain employed, or to
participate in an approved education and training activity.  45 C.F.R. .
255.2(a).  Child care provided or claimed for reimbursement must be
"reasonably related" to the hours of employment or participation in
training, such as the JOBS program.  45 C.F.R. . 255.1(e)(4).  A state
must also provide transportation and other work-related expenses "which
it determines are necessary" to enable an individual to participate in
JOBS activities, and is permitted to provide such transportation and
work related expenses to enable a recipient to participate in approved
education or training in non-JOBS areas.  45 C.F.R. . 255.2(c).

Analysis

Below, we first discuss Oklahoma's challenges to ACF's statistical
sampling methodology.  We then discuss each of the six sample payments
disputed by Oklahoma.

1.      ACF's statistical sampling methodology

The disallowance was calculated using statistical sampling in the
following manner.  First, based on instructions from ACF, Oklahoma
reviewed a sample of child care payments on behalf of individual
children in families receiving AFDC, to determine if they were made in
the correct amount to eligible individuals.  ACF then reviewed a
subsample of those payments; the subsample review identified 15 child
care payments totaling $2,154 in federal funding that ACF or Oklahoma
determined were erroneous.  ACF Exhibit (Ex.) H.  ACF then compared its
error findings with Oklahoma's through a statistical technique called
regression analysis.  This comparison yielded an error rate, or the
percentage of Oklahoma's claim that ACF asserted was paid in error.  The
error rate was applied to Oklahoma's entire claim for child care
payments to determine the amount of the disallowance.

Oklahoma asserted that ACF's statistical methodology was flawed and that
the error rate used to calculate the amount of the disallowance was
therefore invalid.  Oklahoma argued that ACF should have used random
instead of systematic sampling to select the sample of cases to be
reviewed, that ACF's statistical calculations did not have a high enough
level of confidence to be reliable, and that ACF's subsample of cases
was too small and not representative of the sample.  Oklahoma also
asserted that ACF's error rate determination was affected by
"measurement error" (the statistical effect of the parties'
disagreements as to what constituted an erroneous payment for the
purposes of the review), and that ACF had committed an error in
recalculating the disallowance.  As discussed below, we conclude that
Oklahoma's arguments concerning the statistical methodology employed by
ACF lack merit. 2/

Oklahoma also made a comprehensive argument that the federal audit
process was so fraught with methodological and calculation errors that
its results are untrustworthy.  However, we note that ACF made
corrections in its calculations in response to Oklahoma's objections
which it found to be of merit.  Additionally, we discuss below the
remaining statistical flaws asserted by Oklahoma and conclude that none
of them present a sufficient basis to modify or reverse the amended
disallowance further.

a.      The use of systematic sampling

The disallowance was based on Oklahoma's review of a sample of 454
claims, or individual monthly child care payments, and ACF's subsequent
review of a subsample of 76 of those payments.  The claims were selected
through systematic sampling, where an auditor first selects a claim at
random and then selects every nth claim thereafter to achieve the
desired sample size.  In this case, pursuant to ACF's instructions,
Oklahoma first prepared a list of all the payments in its claim for FFP.
The payments were listed in chronological order by month.  Within each
month, the payments were then arranged in order of AFDC case number.
(The case number represented the particular family, or AFDC assistance
unit, for which the payment was made.)  Then, for each case number, or
assistance unit, the payments were listed in order of the "person code"
for each child in the assistance unit for whom the payment was made.
After the payments were arranged in this fashion, the sample was chosen
by using a random start of 205, and a sampling interval of 250.  That
is, the 205th listed payment was the first selected, and then every
250th payment was selected throughout the universe of payments, yielding
a sample of 454 payments.  ACF then selected its federal subsample of 76
payments from Oklahoma's sample by using a random start of four, and a
sampling interval of six.  ACF Brief (Br.), Att. 2, .. 13, 16.

Oklahoma asserted that ACF's use of systematic sampling was contrary to
good statistical practice.  It provided two affidavits from Roger
Lanier, Ph.D., a research analyst with Oklahoma's Research, Evaluation
and Statistics Unit, who asserted that statistical authorities
recognized by ACF warn against the use of systematic sampling when there
is any risk that the data being sampled is arranged in a periodic or
cyclical fashion.  First Lanier Affidavit (Aff.), .. 32-33.  He noted
that here the claims were organized by families and chronologically, and
asserted that there may have been what he characterized as "other
unknown periodicities" in the universe list of cases.  Id.

The Board has in other cases upheld the use of systematic sampling in
the absence of any showing that it introduced a bias into the sample
results.  Mississippi Dept. of Human Services, DAB No. 1267 (1991); New
Mexico Human Services Dept., DAB No 1224 (1991).  One of the authorities
cited by Dr. Lanier, Arkin, says that systematic sampling is recognized
as valid if the universe is designed so that periods or cycles which
could cause bias are avoided.  Arkin, H., Handbook of Sampling for
Auditing and Accounting, (3d ed. 1984) at 40-43, Oklahoma Exhibit (Ex.)
23.  Here, the Lanier affidavit merely notes that the claims were
organized by families and chronologically, and that "there may have been
other unknown periodicities as well."  First Lanier Aff., . 33.
However, Oklahoma did not offer any explanation of how the arrangement
of the universe by month and by case number within each month introduced
any cycles or periods that could have resulted in a biased sample.

Additionally, as noted in New Mexico, many sampling applications,
particularly in auditing, rely heavily on systematic sampling, and
systematic sampling may even have advantages over random sampling in
that it insures inclusion of cases from the beginning, middle and end of
the universe.  New Mexico, 17-18.  Thus, the chronological arrangement
of the payments in question here arguably contributed to the statistical
integrity of ACF's error rate calculation.  Another of the authorities
cited by Oklahoma states that systematic sampling is sometimes used, for
convenience, in populations in which the numbering of the units is
effectively random, such as in sampling from a file arranged
alphabetically by surnames, if the item that is being measured has no
relation to the surname of the individual.  Cochran, W. Sampling
Techniques (3d ed. 1977) at 213.  Here, the claims were arranged by
month, and, within each month, by case number.  There has been no
suggestion that either the eligibility for or the amount of the child
care payment for a case has any relation to either the month of payment
or the case number.  Accordingly, we conclude that ACF's use of
systematic sampling was appropriate and in accordance with accepted
statistical practice.

b.      The size of the confidence interval

Oklahoma asserted that ACF should have calculated the disallowance based
on a 95% confidence interval, rather than the 90% confidence interval
that was used.  Oklahoma argued that a 95% confidence interval is
commonly used in auditing situations, and noted that a 95% interval was
used to calculate the disallowance in Ohio Dept. of Human Services, DAB
No. 1202 (1990), a decision cited by ACF in support of its statistical
methodology.  Oklahoma argued that it should have been given the benefit
of the wider confidence interval.

A "confidence interval" is a statistician's calculation of a range of
values within which the actual value of the error rate is known to fall,
with a specified degree of probability.  The range of values is
expressed in percentage points above and below the estimate of the error
rate yielded by the regression analysis, called the "point estimate."
In this case, ACF calculated the point estimate of the error rate as
14.8484% with a 90% confidence interval of plus or minus 7.9917%.  ACF
Response to Request for Further Briefing, Att.  Thus, there was a 90%
probability that the true error rate was within 7.9917 percentage points
above or below the error rate point estimate of 14.8484%.  To calculate
the disallowance, ACF used the lower bound of the 90% confidence
interval, 6.8567% (14.8484 minus 7.9917).

The Board has previously considered and upheld disallowances based on
the lower bound of a 90% confidence interval.  In Pennsylvania Dept. of
Public Welfare, DAB No. 1278 (1991), the Board noted that ACF did not
base its disallowance on the point estimate of the error rate, but
rather on the lower limit of the two-sided 90% confidence interval
around the point estimate.  A 90% confidence interval means that there
is a 10% probability that the true value of the error rate falls outside
the confidence interval; or a 5% probability that the true value is
greater than the upper limit or bound of the confidence interval, and a
5% probability that it is below the lower limit.  Thus, since the
disallowance was based on the lower limit of the confidence interval,
and not the point estimate, there was a 95% probability that the true
value was above the lower limit.  The Board upheld the disallowance in
Pennsylvania since the state was protected with a 95% degree of
confidence from having to pay an amount greater than the true value of
erroneous payments.  Pennsylvania at 8; see also New York State Dept. of
Social Services, DAB No. 1358 at 45-46 (1992) (using the lower limit of
a 90% confidence interval allowed the auditors to estimate with a 95%
probability that the true value of unallowable payments equalled or
exceeded the recommended disallowance); California Dept. of Social
Services, DAB No. 816, at 5 (1986).  The use of the 90% confidence
interval in the case of Oklahoma is justified on the same basis.

Furthermore, the Board's decision in Ohio does not compel the use of a
95% confidence interval in this case.  In Ohio, the statistical analysis
was used to determine whether the State had provided required services
in 75% or more of child support enforcement cases under title IV-D of
the Act.  Failure to meet the 75% threshold would result in a percentage
reduction of federal program funding until compliance was achieved.
Thus, the statistical methodology was used not to calculate a
disallowance, but to determine whether there would be a penalty
reduction imposed pursuant to a statutory formula.  (A determination
that Ohio failed to meet the threshold triggered a five million dollar
disallowance for the fiscal year in question.)  Moreover, the use of a
95% confidence interval for this purpose in the IV-D program is dictated
by IV-D program policies.  Conversely, in the instant case, statistical
sampling is being used to compute the actual amount of the disallowance
of erroneous payments made under the IV-A program.  As stated above, the
90% confidence interval gave Oklahoma a very high degree of protection
from having to pay more than the true value of erroneous payments.
Accordingly, we conclude that the situation in Ohio was not analogous to
the disallowance here.

c.      Sample size

Oklahoma asserted that the sample size fell far short of ACF's own
specifications for conducting the review of Oklahoma's claim.  It cited
a memorandum from an ACF statistician which stated that the sample
should be "large enough to provide a precision of plus or minus two
percentage points from the mid-point for an estimated .2 attribute at
the 90 percent confidence level for a two tailed test," necessitating a
sample size of 1,083 independent claims, and a federal subsample of 220
independent claims.  Oklahoma Ex. 16.  That is, according to this
statistician, the sample size should be such that there would be a 90%
probability that the actual error rate would fall within a range, or
confidence interval, of plus-or-minus two percentage points from the
error rate yielded by the regression analysis.

The sample of 454 payments, and the subsample of 76 payments, was much
smaller than recommended by the ACF statistician.  However, ACF's
revised calculations accounted for the smaller sample and subsample
sizes with a much wider confidence interval.  Rather than a confidence
interval of plus or minus two percentage points (that would have
resulted from the use of the larger sample and subsample sizes), the
revised calculations were based on a confidence interval of
approximately eight percentage points in either direction.
Additionally, ACF calculated the disallowance using the lower bound of
the confidence interval as the error rate.  Thus, while the error rate
used to determine the disallowance with the larger sample size would
have been approximately two percentage points less than the that yielded
by the regression analysis with the smaller sample and subsample, the
error rate actually used to determine the disallowance amount was almost
eight percentage points less than the point estimate of the error rate.
ACF Ex. G.  In effect, this gives Oklahoma the benefit of any doubt
raised by use of the smaller sample.  Oklahoma did not argue that the
resulting confidence interval failed to appropriately reflect the degree
of precision provided by the smaller sample sizes.  Accordingly, we
conclude that Oklahoma has not shown that the sample and subsample sizes
rendered ACF's calculations unreliable.

We also note that a larger sample and subsample were originally used to
calculate the disallowance, commensurate with the ACF statistician's
recommendation, but that the sample sizes were reduced in response to
arguments presented by Oklahoma.  The disallowance was initially based
on a sample of 1,161 payments, and a subsample of 204 payments.  The
larger sample was selected through the use of "cluster sampling," by
which ACF expanded the sample of 454 payments that Oklahoma initially
selected to include, for each payment selected for a particular child,
all payments made during the same month for the child's siblings as
well.  Similarly, ACF's initial subsample of 76 payments was expanded to
220 payments by including payments for siblings.  However, after
Oklahoma objected to the use of cluster sampling, ACF recalculated the
disallowance using only the originally selected sample without the
addition of the sibling payments.

d.      Representativeness of the subsample and measurement error

Oklahoma asserted that the subsample was unrepresentative of the sample
because the payments Oklahoma identified as erroneous, while comprising
only 2.87% of total child care payments in the sample, comprised 6.98%
of total payments in ACF's subsample. 3/  Oklahoma characterized ACF's
sampling methodology and design as flawed, and asserted that the
differences in the error rates showed that the error rate for the
subsample was biased upwards.  Oklahoma Reply Br. at 3, 6.  However,
Oklahoma did not cite any specific deficiencies in the sample or
subsample selection process which would have caused the difference in
the error rates.  We concur with ACF that some difference may always be
expected, and that the difference here did not by itself demonstrate
that the sample was unreliable. 4/

Oklahoma also noted that when the payments in the subsample that ACF
identified as erroneous (but which Oklahoma considered proper) were also
considered, the federal error rate for the subsample was 20.41%.
Oklahoma attributed the higher rate of ACF-identified erroneous payments
to "measurement error," which Oklahoma characterized as resulting from
the absence of any clear and objective definition of error by ACF, and
differences between the parties as to what constitutes an erroneous
payment.  Simply put, measurement error here refers to Oklahoma's
position that ACF classified as erroneous some payments that Oklahoma
considered correct. 5/

However, we note that Oklahoma has had the opportunity to contest ACF's
error determinations before the Board, and that during the course of
this appeal ACF reversed its error findings with respect to four of the
ten individual payments that were appealed by Oklahoma, and that our
decision reverses four more.  Thus, the actual error rate for the
subsample was considerably less than the 20.41% figure originally cited
by Oklahoma.

Oklahoma also complained that it was not given an opportunity to rebut
ACF's error findings prior to the disallowance, despite being told twice
that they would be able to reconcile all differences of opinion and
interpretation relating to the review findings prior to settling the
claim.  It asserted that this denial of the opportunity to contest ACF's
findings was not made up for by the later opportunity to challenge the
disallowance before the Board, and noted that the passage of time had
hindered its ability to obtain information necessary to contest ACF's
findings.  However, Oklahoma did not cite any program requirement
granting it the right to contest ACF's error determinations prior to the
disallowance being issued.  Nor has Oklahoma cited any specific instance
where the denial of such opportunity prevented it from obtaining
information necessary to prosecute its appeal.  Oklahoma had an ongoing
responsibility to document the allowability of its claims and is
required to retain records for three years or longer if an audit or
other review begins before the three-year period expires.  45 C.F.R.
74.21.  Here, an initial form of sample review began at least by May 3,
1991.  Oklahoma Ex. 7.

Therefore, we conclude that Oklahoma has not shown that it was
prejudiced in its efforts to correct any errors in the disallowance.

We thus uphold ACF's statistical methodology and calculations.

2.      Individual payment error determinations

Oklahoma originally disputed ACF's error findings for ten individual
child care payments in the subsample.  During the course of the appeal,
ACF agreed that it made clerical mistakes in recording the amount of
Oklahoma's error findings for two of the payments, and agreed to exclude
two other payments from the calculation of the disallowance.  ACF Br. at
22; ACF Response to Request for Further Briefing at 2-3.  Thus, we need
consider only six disputed error findings.  For all but one of these six
payments, Oklahoma argued that ACF had incorrectly found the case to be
ineligible for the child care payment.  In the other payment, the first
one discussed below, ACF agreed with Oklahoma that the payment was
correct but treated the case as a State-identified error for the
purposes of the regression analysis.  As explained below, we reverse
ACF's determinations for four of the six payments in dispute and uphold
its determinations for the remaining two payments.

a.      AFDC Case No. 132891, review month August 1990

ACF and Oklahoma agreed on appeal that this was a correct payment to an
eligible recipient.  However, for purposes of its statistical
calculations, ACF treated this as a payment which Oklahoma had found to
be erroneous but which ACF subsequently determined to be correct, since
Oklahoma's child care case review form for this payment showed that it
was not authorized or correct.  Oklahoma Ex. 35.  Accordingly, ACF's
list of the error cases shows a "state error" finding of $10.00 for this
payment, and a "federal error" finding of $0.00.  ACF Ex. H at 1.

ACF argued that Oklahoma's child care case review form amounted to a
finding of an error, even though the space indicating an error may have
been checked by mistake.  ACF reported that in determining an error rate
through regression analysis, all state findings in which there is an
error must be included, including any case which was correctly paid to
an eligible child but which the state has mistakenly called an error
case.  ACF noted that it had found no payment error for this case, and
asserted that the regression formula took into account ACF's finding
that the case was correctly paid, and that Oklahoma was therefore not
penalized. 6/

While ACF noted that the review form had been completed incorrectly, ACF
did not dispute Oklahoma's position that Oklahoma had not found the case
to be ineligible.  Oklahoma reported that the payment did not appear on
Oklahoma's list of error cases, that the review form did not report an
erroneous payment amount or any basis for the supposed error finding,
and that the payment had not been submitted for an additional level of
State review as were other error cases.  First Aff. of CharlesEtta Combs
and Sandra Headrick, .. 19-20, Oklahoma Br., Att. B.  Thus, ACF did not
establish that Oklahoma's review had determined, albeit incorrectly,
that this was an erroneous payment.

We also note that, for the purposes of ACF's statistical computations,
Oklahoma's error findings are expressed in terms of dollars paid in
error.  ACF Exs. G, H.  However, Oklahoma's review form did not report
any dollar amount as having been paid in error for this claim.  Oklahoma
Ex. 35.  Additionally, it appears that ACF corrected federal and State
error payment amounts which resulted from clerical errors.  ACF Br. at
22.

Accordingly, we conclude that Oklahoma did not report an error finding
for this payment.  Since ACF failed to substantiate that Oklahoma had
not been adversely affected by ACF's treatment of this case as a State
finding of a $10 payment error and since ACF had previously corrected
State error payment amounts which resulted from clerical errors (and
since ACF will have to recalculate the disallowance amount in any
event), we reverse ACF's determination that this payment must be treated
as an Oklahoma error determination of $10 in the regression analysis.

b.      AFDC Case No. 052225, review month March 1990

ACF disallowed this child care payment, which Oklahoma authorized to
permit the client to attend school, on the grounds that there was no
evidence that the client actually attended the classes she was
registered for, or that she attended them during the March 1990 review
month.  ACF noted that Oklahoma's documentation, a class registration
form showing the classes in which the client was enrolled for the period
January 10 through May 11, 1990 (the spring semester), was dated
December 27, 1989, and thus did not establish that she had attended
classes in March, 1990.  Oklahoma Ex. 36 at 5.  ACF therefore concluded
that the need and basis for child care was not substantiated, and that
Oklahoma failed to meet its burden of documenting the allowability of
its claim.

In response, Oklahoma noted that the client had registered for English
composition I during the spring semester and English composition II
during the following semester.  Therefore, Oklahoma argued, the client
must have completed English composition I, and thus attended classes
during the review month, since English composition I was a prerequisite
for English composition II.  Oklahoma noted that the December 27, 1989
registration form showed that the client had registered for English
composition I during the spring semester, which included the review
month.  Oklahoma Ex. 36 at 5.  Oklahoma also provided a class schedule
dated April 18, 1990 showing that the client was registered for classes,
including English composition II, for the period June 4 through July 30,
1990.  Oklahoma Ex. 46.

ACF raised a valid objection to this payment, that the class
registration form predated the review month and thus did not establish
the class attendance on which eligibility for child care was based.
However, Oklahoma answered ACF's objection with additional documentation
which reasonably established that the client completed one of the
courses for which she had been registered during the semester which
included the review month.  Accordingly, we conclude that Oklahoma
established that the client had attended this class during the review
month, as opposed to merely having registered for classes, as ACF
asserted, and was thus eligible for the child care payment. 7/

c.      AFDC Case No. 004782, review month November 1989.

ACF asserted that Oklahoma failed to establish that the client was
eligible for full-time child care, which is authorized by Oklahoma's
AFDC State plan when needed for at least four hours per day.  Oklahoma
Ex. 26 at 2.  ACF noted that the client was attending classes for only
three hours a day, and asserted that Oklahoma failed to document its
claim that the client required at least an additional hour to travel to
and from class.  ACF noted that Oklahoma had reported that test trips to
the class took 17 minutes for travel alone and 35-45 minutes including
time required to find parking.  These travel times were inconsistent
with Oklahoma's authorization of half an hour to travel to the class and
one and a half hours for the return trip, ACF asserted, since parking
would not be needed on the return trip.

ACF also noted that Oklahoma had reported that more time would be
required if the client used public transportation, which demonstrated
that Oklahoma did not know what form of transportation the client
utilized to attend classes.  Accordingly, ACF asserted, Oklahoma's
authorization of full-time child care was not reasonably related to the
client's needs, since Oklahoma did not know what those needs were.

ACF did not take issue with Oklahoma's position that its State plan
permitted payment for full-time child care when child care is needed for
at least four hours per day, and agreed that travel time could be
included in determining the amount of time for which child care is
needed.  Accordingly, Oklahoma would only have to demonstrate that the
client required at least one hour of travel time per day in order to
qualify for full-time child care under its State plan.  Even accepting
ACF's position that parking would not be required for the return trip,
Oklahoma demonstrated the need for 52 to 62 minutes travel time per day,
based on its undisputed test times of 17 minutes travel time without
parking, plus 35-45 minutes including parking.  Thus, Oklahoma
established that travel time could reasonably require at least one hour
per day.  Additionally, Oklahoma argued that the client would have to
find parking on the return trip, since she would have to stop to pick up
her child from day care.  Oklahoma further reported that additional time
had been granted for the afternoon trip to account for the possibilities
of late-afternoon traffic and classes running overtime.  Oklahoma also
noted that it had provided information on public transportation only to
anticipate any arguments that the client could have used public
transportation to avoid parking problems.

Again, this is a situation where ACF raised objections to Oklahoma's
claim, and Oklahoma provided a reasonable explanation which, in the
Board's view, satisfied the objections and fulfilled Oklahoma's burden
of documenting the allowability of the expenditure.  Accordingly, we
conclude that Oklahoma demonstrated the client's entitlement to
full-time child care under its AFDC State plan, and we reverse ACF's
disallowance of this claim.

d.      AFDC Case No. 069834, review month November 1989

ACF disallowed this payment on the ground that it was not made on behalf
of the child for whom child care was authorized.  ACF asserted that
while care had been authorized for the client's five-year-old daughter
Alycia, the actual payment had been made not for Alycia, but for her
seven-year-old sister, Alexia.  ACF cited an Oklahoma computer form
generated for the review of Oklahoma's claim for FFP which reports that
payment had been made for "Ale, 7" which, ACF said, referred to Alexia.

In response, Oklahoma presented evidence to show that Alycia was the
child who received the authorized day care services and for whom payment
was claimed by the child care provider.  This evidence included:

o       A child care authorization form, authorizing child care for
Alycia, date of birth October 19, 1984.  Oklahoma Ex. 38 at 4.  That
form is directed to "Day School #2."

o       A notation in the case narrative indicating that Alycia would be
attending day school #2.  Id. at 6.

o       A claim form from "Day School #2" on behalf of "Alisha," date of
birth October 19, 1984, reflecting attendance during the review month of
November 1989.  Oklahoma Ex. 47.

Oklahoma asserted that the computer-generated review form produced for
the audit of its claim, which showed the payment as having been made on
behalf of Alexia, was erroneous.  Oklahoma Ex. 38 at 3.  Oklahoma
explained that the day care center had mistakenly used the "person
number" code for Alexia when it claimed payment for Alycia, and that
Oklahoma's financial system therefore recorded the day care claim in
Alexia's name and generated a review form for her instead of Alycia.

An Oklahoma case member list for the family shows Alexia's "MB" number
was "M12", and Alycia's "M13".  Oklahoma Ex. 48.  However, the day care
provider's claim form, which shows that services were provided for
Alisha, date of birth October 19, 1984, lists her person code as "12".
Oklahoma Ex. 47.  Thus, Oklahoma's claim, that Alycia was the child
attending care and thus was the child for whom payment was made, appears
reasonable.  Accordingly, we conclude that Oklahoma reasonably
demonstrated the allowability of this payment, and we reverse ACF's
disallowance. 8/

e.      AFDC Case No. 039540, Review month May 1990; and AFDC Case No.
008265, Review month June 1990.

ACF disallowed these payments because in each case child care had been
authorized to permit the employment of a custodial relative (a maternal
aunt and a grandmother) who was not included in the AFDC assistance
unit, and thus not a recipient of AFDC.

Oklahoma argued that the custodial relatives were eligible for these
payments because the plain language of section 402(g) of the Act
requires states to guarantee child care to permit "an individual in the
family" to be employed or attend specified training.  Section
402(g)(1)(A) of the Act.  However, as Oklahoma acknowledged, the
applicable regulation requires states to provide AFDC child care only
when necessary to permit "an AFDC eligible" family member to work or
attend training.  45 C.F.R. . 255.2(a).  Here, Oklahoma did not argue
that the custodial relatives were eligible for AFDC, and reported that
they either chose to exclude themselves from the AFDC assistance units
or exceeded applicable financial limits.  Thus, under the plain language
of the regulation, they were not eligible for the child care payments.

Oklahoma asserted, however, that ACF has interpreted the regulation too
narrowly and in a manner inconsistent with statute's plain language and
its purpose, as evidenced by the Family Support Act's legislative
history, of reducing dependence on welfare.  Oklahoma noted that the
implementing regulation is preceded by a statement that child care is to
be provided for families receiving AFDC.  45 C.F.R. . 255.0.  Only the
most compelling reason, Oklahoma argued, would justify upholding an ACF
interpretation that is different from the statutory language and
contrary to the statutory goals.

The Board gives deference to an agency's interpretation in regulations
of the statutes that it administers, and in fact is bound by all
applicable regulations.  Virginia Dept. of Health, DAB No. 208 (1981);
45 C.F.R. . 16.14.

Moreover, Oklahoma did not persuade us that the regulation here is
inconsistent with the language and goals of the statute.  Although
section 402(g) of the Act authorizes child care to permit "an individual
in the family" to work or attend training, Oklahoma has not shown that
the term "family" as used in section 402(g) or in other provisions of
the Act would include individuals who are not part of the AFDC
assistance unit.  The general statements of purpose in the legislative
history that Oklahoma cited do not establish that the term "family" in
the section 402(g) of the Act (added by the Family Support Act) was
intended to have this broad a meaning.  Accordingly, we reject
Oklahoma's argument on statutory intent.

Moreover, as ACF noted, its interpretation of the 402(g) child care
provisions is consistent with the disregard of child care costs under
402(a)(8)(A)(iii) of the Act, which applies only to the earned income of
an individual whose income and needs are included in determining
eligibility for AFDC.  Under both of these provisions of the Act, the
costs of providing child care may only be considered in relation to
income that is actually available to the assistance unit.  Both
provisions make funds available to the assistance unit:  section 402(g)
makes funds directly available by providing child care payments, and
section 402(a)(8)(A)(iii) makes funds available indirectly by
disregarding income attributable to the cost of child care in
determining need, income that would otherwise be considered available to
the assistance unit.  In both cases, funds are available only where the
income of the individual bearing the cost of the child care is
considered available to the assistance unit.  By contrast, under
Oklahoma's broad interpretation of section 402(g) of the Act, funds for
child care would be made available to an individual, the custodial
relative, whose income is not available to the assistance unit.  The
disregard provision further shows that ACF's interpretation is
consistent with the Act.

Oklahoma also argued that the regulation does not prohibit a state from
providing child care to permit a non-AFDC eligible relative to work or
attend training.  However, Oklahoma cited no specific provision of the
regulation requiring ACF to provide FFP in child care expenditures other
than those authorized by section 255.2.  In this regard, we note that
the regulation does specify other costs in which the state may receive
FFP at its option, such as reimbursement for transportation and other
work-related expenses in approved non-JOBS training.  45 C.F.R. .
255.2(c)(2)(i).  Thus, it can be inferred from the absence of any
similar provision permitting states to provide child care for non-AFDC
eligible family members that FFP on such costs is not available.

Accordingly, we sustain ACF's disallowance of these two child care
payments.

Conclusion

As explained above, we uphold ACF's use of statistical sampling in
computing the disallowance.  We sustain the disallowance with respect to
AFDC Case Nos. 039540, 008265.  We reverse the disallowance with respect
to AFDC Case Nos. 132891, 052225, 004782, and 069834.  ACF should
recalculate the disallowance accordingly and notify Oklahoma of the
revised disallowance amount.  If Oklahoma disagrees with ACF's
recalculation, it may return to the Board for review of this limited
issue within 30 days of receipt of ACF's determination.

   M. Terry Johnson

   Judith A. Ballard

  Donald F. Garrett Presiding Board Member

1.   The original disallowance totaled $1,471,499.  ACF reduced the
amount during the course of these proceedings in response to arguments
and evidence provided by Oklahoma.

2.   Oklahoma also asserted that ACF's recalculation of the disallowance
was flawed and overstated the disallowance by several thousand dollars
because it accounted "only for the sampling variability of the numerator
of the fraction that comprises the error rate."  Oklahoma argued that
"the sampling variability in the denominator was ignored. . . .  This
error should have been avoided by substituting the actual population
value in place of the sample-based value in the denominator."  Oklahoma
Reply Brief (Br.) at 4.  ACF disagreed and stated that "(t)he regression
error for the payment error rate is a composite estimator in which the
numerator is the regressed average error amount paid per case, and the
denominator is the average payment paid per case, as computed from the
State full sample."  ACF Response to Board Request for Further Briefing
at 2 (emphasis in original).  Thus, ACF asserted, the entire fraction
was appropriately sample based.  The Board provided Oklahoma an
opportunity to comment on ACF's response, and stated that if Oklahoma
continued to dispute ACF on this point, then it should support its
position, that the actual population value should have been used in the
denominator, with citations to appropriate statistical authorities.  In
response, Oklahoma reported that in view of the small amount involved,
it had decided not to pursue this point any further at this time.
Oklahoma Response to Request for Comments.  Accordingly, we conclude
that Oklahoma failed to establish that ACF committed an error in the
recalculation.

3.   The error rates cited by the parties in their initial briefs were
based on the larger sample and subsample that included sibling payments
selected through cluster sampling.  The parties' submissions did not
provide specific rates of Oklahoma- and ACF-identified errors for the
smaller sample and subsample sizes which ACF utilized to calculate the
revised disallowance amount.

4.   Oklahoma also asserted that much of the difference between the
rates of State-identified errors in the sample and in ACF's subsample
was attributable to ACF's use of cluster sampling.  However, as noted
above, ACF did not use cluster sampling when it recalculated the
disallowance during the appeal.

5.   In New York State Dept. of Social Services, DAB No. 522 (1984), the
Board noted that the term "measurement error," as used by the parties to
that appeal, referred to errors in the universe which may be
attributable to, for example, tabulation or sorting errors, wrong
answers, and, generally, erroneous classification of material.  New York
at 90, fn. 39.

6.   ACF's statement that Oklahoma would not be penalized is ambiguous,
since ACF persists here in disputing Oklahoma's position and since ACF
did not provide any corroborating evidence concerning the impact of
Oklahoma's position on any recalculation.

7.   With respect to the payment for this case and two others (Nos.
004782 and 069834), Oklahoma argued that ACF had not issued specific
instructions regarding what documentation must be in a case record to
support a claim for child care.  Since we conclude that Oklahoma
provided sufficient evidence on appeal to answer ACF's questions
concerning the payments and satisfy its burden of establishing that they
were allowable under the Act, the regulations, and Oklahoma's AFDC State
plan, it is not necessary to address the question of whether the absence
of instructions regarding documentation caused Oklahoma any harm.

8.    We note that the "mistake" in using the wrong person code for
Alycia appears to have been committed by Oklahoma, and not by the day
care center.  Oklahoma's child care authorization form directed to the
day care center, which authorized child care for Alycia, lists "12" as
her person code, and it appears that this information was used by the
day care center in preparing its claim.  However, this does not alter
the fact that the record demonstrates that Alycia was the child who
actually received the authorized