Quiero saber acerca de...

khammond on DSKJM1Z7X2PROD with RULES2

Federal Register / Vol. 86, No. 142 / Wednesday, July 28, 2021 / Rules and Regulations selects control variables in two stages: The first stage runs a Lasso regressing the dependent variable on a set of common controls; the second stage regresses the explanatory variables of interest on the same set of common controls. A simple Lasso only selects predictors, without the possibility of statistical inference afforded by double selection. The Commission focuses on GTL, ICSolutions, and Securus because these firms costs explain the bulk of industry costs. These providers supply 58% of all inmate calling services contracts, and cover approximately 78% of all incarcerated people as measured by average daily population.
These shares may in fact represent an understatement of their industry share because, for example, CenturyLink, a large provider when judged by average daily population, subcontracts almost all of its contracts to ICSolutions, and, in the case of the large Texas Department of Corrections contract, to Securus. These three firms are also more suitable for making cross-firm comparisons because they do not subcontract the provision of inmate calling services to a third party, and because they are the largest three of the five providers that serve prisons, covering 111or 86%of all prison contracts. Of the remaining prison providers, CenturyLink supplies REDACTED prison contracts, Legacy supplies REDACTED, and NCIC supplies REDACTED. The results illustrate how high GTLs reported costs are relative to those of its nearest peers, showing GTLs costs to beall other things being equalREDACTED greater than the costs reported by Securus and REDACTED greater than the costs reported by ICSolutions. These cost differences are statistically significant at confidence levels greater than 99%. When the sample is restricted to the contracts with no missing rurality information, GTLs costs areall other things being equal approximately REDACTED greater than the costs reported by Securus, and REDACTED
greater than the costs reported by ICSolutions.
7. The results of the double-selection Lasso model also indicate thatall other things being equalthe costs of providing inmate calling services are approximately 22%
greater in jails than in prisons; this difference is statistically significant at confidence levels greater than 99%. For the sample restricted to contracts with complete rurality information, this estimate is approximately 21% and significant at the 99% level of confidence.
8. The Lasso model allows the Commission to consider how a wide array of variables affect a contracts per-minute cost. However, the limitations of the available data may cause the Lasso model to understate the impact of certain variables. For example, because reported costs vary greatly across providers, Lasso may be under-ascribing importance to other variables such as size and type of facility. Commenters criticized the Commissions analysis of reported costs in the 2020 ICS FNPRM. In addition to critiquing the shortcomings of the data used, commenters disagreed with the notion that costs were similar across facility type and size. Some commenters argued that prisons should be expected to have lower per-unit
VerDate Sep<11>2014

19:16 Jul 27, 2021

Jkt 253001

costs than jails, and that larger jails should have lower per-unit costs than jails with average daily populations less than 1,000.
Given the concerns that differences in provider data filing practices impede the Lassos ability to capture the significance of other variables, as well as the economic rationale for the presence of economies of scale in this market, the Commission finds these arguments to be persuasive. The Commission performs additional analyses to investigate differences in cross-provider costs in Appendix C. The approach the Commission uses there attempts to address provider-level cost differences that obscure the relationship between variables such as facility size and a contracts cost.

Appendix C
Lower Bound Analysis 1. Given deficiencies of the cost data submitted by providers, the removal of invalid, incomplete, and otherwise anomalous contracts performed in Appendix A is a necessary step towards determining accurate per-minute costs. In this Appendix, the Commission frequently refers to inmate calling services providers by short names or acronyms. These providers are: ATN, Inc.
ATN; CenturyLink Public Communications, Inc. CenturyLink; Correct Solutions, LLC
Correct; Combined Public Communications CPC; Crown Correctional Telephone, Inc.
Crown; Global TelLink Corporation GTL;
ICSolutions, LLC ICSolutions; Legacy Long Distance International, Inc. Legacy; NCIC
Inmate Communications NCIC; Pay Tel Communications, Inc. Pay Tel; Prodigy Solutions, Inc. Prodigy; and Securus Technologies, LLC Securus. Using those data, the Commission then develops the upper bounds of the zones of reasonableness for the interim interstate provider-related rate caps based on a mean plus one standard deviation approach. However, the upper bounds overstate true per-minute costs by substantial margins. In addition to generally applicable grounds for overstatement, each upper bounds construction includes a number of contracts that the Commission identifies as statistical outliers, and includes all GTL contract costs as reported, despite abundant indicia that GTLs reported costs are both unreliable as a measure of GTLs actual costs of providing inmate calling services and significantly higher than its true costs.
2. In the following analysis, the Commission makes further adjustments to the submitted cost data using generally accepted statistical and econometric techniques. The Commission begins by performing an analysis of statistical outliers to determine whether certain remaining contracts in the data are well outside of the mean of per-minute costs and remove those observations revealed to be outliers by the use of these metrics. Next, the Commission performs a cost adjustment of GTLs reported per-minute contract costs, using reliable information reported for GTLs own contracts as well as the contract information of other inmate calling services providers to identify surrogate observations to use instead of GTLs reported per-minute costs. The results of this analysis allow the Commission to
PO 00000

Frm 00069

Fmt 4701

Sfmt 4700

40749

derive lower bounds of per-minute contract costs for prisons and larger jails. They additionally allow the Commission to address concerns raised in the record regarding expected differences in contract costs across facilities of different types and sizes.
1. Analysis of Outliers 3. As the Commission reviews in detail in Appendix A, the Commission performs an initial round of data cleaning on the contractlevel dataset derived from the Second Mandatory Data Collection by removing contracts with invalid or incomplete data, excluding anomalous contracts, and making additional data adjustments. The final dataset contains 2,900 contract-level observations and is the starting point for the outlier analysis presented here. The Commission now turns to outlier detection and removal.
Using conservative thresholds for both parametric and non-parametric outlier detection techniques that is, techniques that rely on normality assumptions about the distribution of the cost data versus techniques that do not, the Commission finds and removes the data points that are well outside of the central tendency of the distribution of per-minute costs as measured by the mean and standard deviation.
4. The Commission first employs two closely related parametric techniques: The Grubbs test and the modified Thompson Tau test. Both tests detect the largest absolute deviations from the mean divided by the standard deviation. For each approach, if the data point with the largest deviation is above a critical threshold then it is considered an outlier and removed. Both tests continue to iterate through the dataset, recalculating the test statistic and comparing it to the critical value until they no longer detect any outlying observations. The critical regions for the Grubbs and Thompson Tau tests are similar but are based on a different version of the Students t test statistic. For the Grubbs test, the Students t is based on N2 degrees of freedom and a tail value equal to a/2N. For the Thompson Tau test, the Students t is based on N2 degrees of freedom and a tail value of a/2. This difference results in the Thompson Tau test always calculating a lower test statistic than the Grubbs, leading to the detection of more outliers at a given confidence level but also a higher likelihood of false positives.
5. The Commission performs this analysis on the average cost per minute for each contract, and separately for prisons, larger jails, and jails with average daily populations of less than 1,000. The contract-level cost per minute is defined as: contract direct costs +
contract allocated overhead costs/contract total paid minutes. Larger jails have average daily populations greater than or equal to 1,000. As in Appendix A, jails with average daily populations of less than 1,000 are included in the totals to ensure that the Commissions outlier detection and removal is comprehensive among the total dataset of 2,900 contracts. But, because the Commission does not address such jails in the Report and Order for purposes of arriving at interim provider-related rate caps based on the Second Mandatory Data Collection, the
E:FRFM28JYR2.SGM

28JYR2