Employee choice of flexible spending account participation and health plan.
ABSTRACT Despite the fact that flexible spending accounts (FSAs) are becoming an increasingly popular employer-provided health benefit, there has been very little empirical study of FSA use among employees at the individual level. This study contributes to the literature on FSAs using a unique data set that provides three years of employee-level-matched benefits data. Motivated by the theoretical model of FSA choice presented in Cardon and Showalter (J. Health Econ. 2001; 20(6):935-954), we examine the determinants of FSA participation and contribution levels using cross-sectional and random-effect two-part models. FSA participation and health plan choice are also modeled jointly in each year using conditional logit models. We find that, even after controlling for a number of other demographic characteristics, non-whites are less likely to participate in the FSA program, have lower contributions conditional on participation, and have a lower probability of switching to new lower cost share, higher premium plans when they were introduced. We also find evidence that choosing health plans with more expected out-of-pocket expenses is correlated with participation in the FSA program.
- SourceAvailable from: soa.org[Show abstract] [Hide abstract]
ABSTRACT: Models used to derive optimal contributions to health care flexible spending accounts (FSAs) typically assume an employee's household annual out-of-pocket health care expenses are an ab-solutely continuously random variable. This assumption, however, ignores the fact that some em-ployees may be able to accurately predict a portion of their household annual out-of-pocket health care expenses and often actually incur only those expenses during the plan year, implying that a mixed random variable may be more appropriate. In addition, data have shown that employees are setting contributions at lower levels than existing absolutely continuous models would suggest is optimal. Using a mixed model of household annual out-of-pocket health care expenses we prove that it is often optimal for employees to contribute an amount equal to their household annual predictable out-of-pocket expenses, thus avoiding the risk of forfeiture. We also propose a practical rule of thumb that employees may use for setting their FSA contributions. Overall, we recommend that employees use their FSAs to cover only their highly predictable out-of-pocket health care expenses rather than use their FSAs as a contingency fund to pay for unlikely or unexpected out-of-pocket health care expenses.North American Actuarial Journal 07/2011; 15(3).
- [Show abstract] [Hide abstract]
ABSTRACT: I model the interaction of flexible spending accounts (FSAs) and conventional insurance. I show that FSA participation reduces the desired level of insurance coverage. I also show that FSA participation can reduce the total tax cost of health insurance premium and FSA contribution exclusions.The Geneva Risk and Insurance Review 09/2011; 37(2). · 0.63 Impact Factor
- [Show abstract] [Hide abstract]
ABSTRACT: Flexible spending accounts (FSAs) are an important component of health care financing in the United States. Based on a national survey of employer-sponsored health plans, 83 percent of large employers offer flexible spending accounts for health expenses (Mercer, 2008). The average annual contribution amount for those who use FSAs was $1,385 in 2008. Based on 2004 data, the Congressional Budget Office estimated that about 10 million private-sector workers used a FSA for out-of-pocket health expenses (CBO, 2006). The government does not produce a tax expenditure amount for health FSAs alone, but it is likely in the range of $7 billion to $12 billion for 2008.Flexible spending accounts are usually provided through employer “cafeteria plans.” Account holders select an amount to be withheld from their paycheck over the upcoming year. Those monies are put into a FSA and the account holder can use those funds to pay for out-of-pocket health care costs. The key advantage of these accounts is that they escape both income and payroll taxes. But any monies left in the account at the end of the period revert to the employer, generally known as the ‘use-it-or-lose-it’ provision.The use-it-or-lose-it provision adds a substantial amount of risk to the usage of FSAs. Forfeitures could exceed the tax savings, especially for low-income users of FSAs. The Congressional Research Service reports that approximately 1 in 8 FSAs for federal employees forfeited some amount in the first year FSAs were offered (2003), with an average forfeiture of $220 (CRS, 2005). In 2005, the Treasury Department made a major modification to the use-it-or-lose-it provision: account holders are now allowed a ‘grace’ period for using FSA funds. Account holders whose benefit period is based on the calendar year are now allowed to use those funds to pay for expenditures through March 15 of the following year. This means that if an account holder is has excess funds in their account at year end, the account holder can reduce their election amount for the upcoming year by the amount of the excess. This substantially reduces the forfeiture risk.Our research uses a unique panel dataset from a medium sized benefits firm to explore how consumers use FSAs over time. In particular, we test the effect of the new ruling on FSA usage. To our knowledge, this is the first panel data study of flexible spending accounts. The data includes insurance claim information, and FSA election and usage information for approximately 20,000 employees (plus dependents) of about 50 firms from 1998 to 2007. The data also includes income, gender, age, marital status, residence state, household size, and tenure at job. Preliminary results show little effect of the grace period on election amounts and expenditure patterns.Applied Economics 06/2010; · 0.46 Impact Factor
ANDREW YOUNG SCHOOL
O F P O L I C Y S T U D I E S
Andrew Young School of Policy Studies
Research Paper Series
Employee Choice of Flexible
Participation and Health Plan
Barton H. Hamilton
Washington University, St. Louis
Georgia State University
Department of Economics
W. J. Usery Workplace Research Group
This paper can be downloaded at:
The Social Science Research Network Electronic Paper Collection:
Working Paper 08-23
ANDREW YOUNG SCHOOL
O F P O L I C Y S T U D I E S
W.J. Usery Workplace Research Group Paper Series
Working Paper 2008-1-4
Employee Choice of Flexible
Participation and Health Plans
Washington University, St. Louis
Georgia State University
This paper can be downloaded at: http://aysps.gsu.edu/usery/Papers.html
Employee Choice of Flexible Spending Account
Participation and Health Plan
Barton H. Hamilton
John M. Olin School of Business
Washington University in St. Louis
Department of Economics
Andrew Young School of Policy Studies
Georgia State University
Despite the fact that flexible spending accounts (FSAs) are becoming an increasingly popular
employer-provided health benefit, there has been very little empirical study of FSA use among
employees at the individual level. This study contributes to the literature on FSAs through the
use of a unique dataset that provides three years of employee-level matched benefits data.
Motivated by the theoretical model of FSA choice presented in Cardon and Showalter (2001),
we examine the determinants of FSA participation and contribution levels using cross sectional
and random effect two part models. FSA participation and health plan choice are also modeled
jointly in each year using conditional logit models. We find that, even after controlling for a
number of other demographic characteristics, non-whites are less likely to participate in the FSA
program, have lower contributions conditional on participation, and have a lower probability of
switching to new lower cost share, higher premium plans when they were introduced. We also
find evidence that choosing health plans with more expected out-of-pocket expenses is correlated
with use of the FSA program.
JEL classification: I18; J33; H24
* corresponding author address: Department of Economics, Andrew Young School of Policy Studies, Georgia State
University, P.O. Box 3992, Atlanta, GA 30302-3992; telephone: 404 - 413 - 0256; fax: 404 - 413 - 0145; email:
email@example.com. We would like to thank J.S. Butler for his valuable advice, Arne Hole, two anonymous referees
for their comments, and the human resources office at the university we studied for their help with the data. The
authors retain responsibility for any errors.
A portion of James Marton’s time spent on this research was funded through mini-grant 06-104-05 from the W.E.
Upjohn Institute for Employment Research. Portions of this research were completed while Marton was a faculty
member in the Martin School of Public Policy and Administration at the University of Kentucky and a post-doctoral
research fellow at the Taubman Center for Public Policy at Brown University.
Flexible Spending Accounts; Health Insurance; Health Plan Choice
There is a small but growing literature on the use of flexible spending accounts
(FSAs) to finance health (and child care) expenses with pre-tax dollars. FSA
contributions are not subject to federal, state, or local income taxes. In addition,
employers do not pay FICA taxes on FSA contributions. This special tax treatment of
FSAs comes from a 1978 extension of Section 125 of the Internal Revenue Code,
allowing it to apply to certain employee out-of-pocket expenses. Employees in firms
offering a FSA can choose during open enrollment how much of their salary to contribute
to their FSA. This contribution is deducted from the employee’s salary on a pre-tax
basis, typically in equal monthly amounts. The employee can then submit approved
medical expenses for reimbursement from their FSA. One potentially negative feature of
FSAs is that any unused funds are considered forfeited at the end of the year and are
given back to the employer. Recently, the U.S. Department of Treasury has ruled that
employers who offer FSAs under a cafeteria plan can extend the FSA plan year for an
additional 2 ½ months, providing a “grace period” for employees to spend down their
As health care costs continue to rise, FSAs are becoming an increasingly common
work related benefit offered to help employees finance out-of-pocket expenditures.
According to the Kaiser Family Foundation’s 2003 Employer Health Benefits Annual
Survey, 83 percent of “jumbo” firms (firms with 5,000 or more employees) offered FSAs
in 2003, which is an increase over the 69 percent of jumbo firms offering FSAs in 1999.
The tax expenditure from cafeteria plans, to which FSA contributions can be made, is
estimated to be over 23.6 billion dollars in 2005 (Joint Committee on Taxation, 2005).
In order to determine the impact of FSAs on health care consumption and health
coverage, many important questions need to be addressed. Which employees are using
FSAs? Given that an employee chooses to use a FSA, what is the appropriate
contribution level when faced with uncertain health care expenses and the risk of
forfeiture? Because FSA participation and health plan choice are typically determined
simultaneously, what factors influence this joint choice? Besides being of interest in its
own right, having a better understanding of employee utilization of FSAs may also shed
light on how well the “cousin” of the FSA, the Health Savings Account (HSA), performs
as a component of a consumer-driven health care plan.2
There is a small literature on consumer and firm behavior with respect to FSAs
(Cardon and Showalter, 2001; Cardon and Showalter, 2003; Cardon and Showalter
2007). These studies theoretically model the choice of FSA contribution level in a
variety of ways. They show that when employees face a fixed health care expense, the
optimal contribution is increasing in the employee’s marginal tax rate, the employee’s
probability of the health care expense, and the size of the health care expense. In
addition, when the health care expense is drawn from a continuous distribution, the
optimal FSA contribution is increasing in the employee’s level of risk aversion. Another
study models the optimal contribution level and then uses simulations to estimate ranges
of optimal contributions based on different marginal tax rates and price elasticities
(Bhattacharya et al., 2002).
2 Unlike an FSA, that can be used with any health plan or even in the absence of a health plan, an HSA is a
specific component of a “consumer-driven” health plan. Such a plan typically offers a high deductible
along with a pre-funded HSA to help cover the out-of-pocket expenses incurred before exceeding the
deductible. Unlike an FSA, HSA balances can be rolled over from one year to the next, subject to some
limits. Despite these differences, both the FSA and the HSA provide incentives for consumers to be more
cost conscious and both require some sort of reimbursement procedure. For more discussion see Cardon
and Showalter (2007).
Because of the difficulty associated with obtaining consumer-level data on FSA
use, the empirical literature on FSAs has lagged behind the theoretical literature. Cardon
and Showalter (2001) provide perhaps the first empirical study that analyzes consumer-
level data on FSAs. They estimate FSA participation regressions using a cross-section of
1996 calendar year data from a medium sized insurance company. The regressions
suggest that age and family size increase FSA participation at a decreasing rate. In
addition, women, state income tax payers, and higher income individuals are more likely
to participate in a FSA program. Contribution levels, conditional on participation, also
increase with income, and are higher for married individuals and those with larger
Another recent study uses data from the cross-sectional 1993 Robert Wood
Johnson Employer Health Insurance Survey (EHIS) to examine the relationship between
employee FSA use and the level of cost sharing associated with the employee’s insurance
plan (Jack et al., 2006). When health insurance is offered by employers in conjunction
with a FSA, their results suggest that the coinsurance rate associated with the health
insurance plan is higher. They claim the resulting increase in out-of-pocket spending
may be efficiency enhancing, as it may potentially help offset the distortionary effect of
the tax subsidy given to employer-provided health insurance premiums.
Our paper extends the empirical literature on FSA use by taking advantage of a
unique data set consisting of three recent years of matched human resources data for
employees of a large public university in the southeast.3 The dataset provides detailed
information (e.g., prices, deductibles) on the set of health insurance and FSA options
available to workers at this employer in each year. This type of detailed plan information
3 Examples of other papers that utilize human resources data from a single firm to examine health benefit
choices include Cutler and Reber (1998) and Parente et al. (2004).
is often unavailable or incomplete in large scale random population surveys. In addition,
we have other explanatory variables such as race and job tenure that have not been
previously analyzed in the literature.
Using the theoretical model of consumer FSA choice presented in Cardon and
Showalter (2001) as our motivation, we estimate two part models of participation and
contribution levels to investigate the impact of demographic and financial characteristics
on these choices. In addition, we recognize that in each year the employee is faced with a
number of health benefit “bundles” that consist of the joint choice of an insurance plan
and whether or not to participate in the FSA program.4 Consequently, we estimate a
conditional logit model that formalizes the idea that insurance plan choice and the choice
to participate in the FSA program are typically made simultaneously. Finally, we use a
multinomial logit (MNL) model to examine employee sorting into new plans with lower
deductibles that were initially offered in year two of our panel to investigate the
relationship between employee characteristics and cost sharing.
As predicted by the theoretical model, we find that FSA participation is positively
correlated with employee marginal tax rates, proxies for health type such as being a
female or older worker, and a proxy for the amount of health care expenses (being
married). Choice of health plans with more cost sharing is also correlated with increased
FSA participation. FSA contribution levels increase with income, age, and for married
employees. A notable finding is that non-whites are less likely to participate in the FSA
program, and make smaller contributions conditional on participation, even after
controlling for other factors such as income. When FSA participation and health plan
4 For example, in 2001, employees can choose between an HMO, a PPO, or no coverage. By combining
these three health plan options with the yes or no FSA participation decision each employee has six health
benefit “bundles” to choose from in 2001.
choice is viewed as a joint decision, age, salary, and marital status are correlated with the
choice of a “bundle” that includes a positive FSA contribution. Finally, our analysis of
employee sorting across new health plans suggests that proxies for health type (age,
gender), a proxy for the amount of health expenses (marital status), income, and previous
plan choice have a statistically significant association with the choice to switch to one of
the new lower cost share / risk, higher premium plans. Even after controlling for a
number of other factors, non-whites have a lower probability of switching to one of these
new lower risk plans.
Theoretical Model of Cardon and Showalter (2001)
In this section, we briefly describe a theoretical model of FSA choice presented in
Cardon and Showalter (2001), which provides motivation for the empirical models we
estimate in this paper. Assume that with probability q an employee (or their family) faces
a potential loss of $L due to health care expenses. This loss may be a function of a
variety of measurable characteristics of the employee or their family, including their level
of insurance coverage and family size. The probability of incurring a loss due to health
care expenses is also a function of a variety of measurable characteristics including age,
gender, race, and education level. Let I be the employee’s pre-tax income, τ be their
marginal tax rate, and F be their choice of FSA contribution level, then the employee’s
state contingent non-health expenditures are given by M:
State of the World Probability Non-health Expenditures
Mloss = (I – F) * (1 – τ) – L + F
No Loss (1 – q)
Mno loss = (I – F) * (1 – τ)
Given these state contingent budget constraints, the employee chooses F to maximize
their expected utility:
choose F to maximize EU(F) = q * U(Mloss) + (1 – q) * U(Mno loss)
Allowing employees to choose their FSA contribution level (as opposed to
offering a fixed contribution level) is supported in theory by models of benefit provision
that assume imperfect worker sorting across firms (Dranove et al., 2000; Gruber and
McKnight, 2003; Levy, 1998). This imperfect sorting implies heterogeneity in tastes for
insurance and FSA contributions. As in Cardon and Showalter (2001), utility is assumed
to exhibit decreasing marginal utility, risk aversion, and decreasing absolute risk aversion
(DARA).5 Because the optimization problem is globally concave in F, there is a unique
optimal FSA contribution level, F*. Focusing on an initial interior solution (0 < F* < L),
the comparative static results can be summarized as follows:
• as q ↑, F* ↑
• as I ↑, F* ↓
• as L ↑, F* ↑
• as τ ↑, F* ↑
In addition, Cardon and Showalter (2001) present a simulation that predicts that
increasing levels of risk aversion lead to greater FSA contributions.
The intuition behind these results is given through a comparison of the expected
marginal benefit of the FSA in the loss state and the expected marginal cost of the FSA in
the no loss state. For example, if the loss due to illness (L) increases, the expected
marginal benefit of the FSA in the loss state increases, while the expected marginal cost
in the no loss state stays the same. This implies an increase in the optimal FSA
5 These assumptions imply U′ > 0, U′′ < 0, and U′′′ > (U′′)2 / U′ > 0. The DARA assumption implies that as
wealth increases, an individual is willing to bear more risk. It is made to sign otherwise ambiguous
comparative static results. See Cardon and Showalter (2001) for more discussion.
contribution (F*). For a more complete description of the comparative static results, see
Cardon and Showalter (2001).
In order to bring the theoretical model to the data, employee income must be
measured and their marginal tax rate estimated. In addition, proxies for the probability of
a health loss, the dollar value of a health loss, and employee risk aversion must be found.
As is discussed in more detail below, the human resources data used in this paper
includes employee income (I). Using employee demographic characteristics, their
marginal tax rate (τ) is estimated using the NBER’s TAXSIM tax calculator. While we
have no direct measures of health status or health claims, employee characteristics such
as age, gender, race, and education level (as measured by a faculty vs. staff indicator) are
used as proxies for the probability of having a health care expense (q) or health type.
Holding these characteristics constant, we use type of insurance and marital status as
proxies for the size of the health care expense (L). Unfortunately, as is the case for most
data sets, direct measures of employee risk aversion are not available. However, it may
be the case that more risk averse employees are less likely to leave the university, after
controlling for age and income (which presumably reflects the worker’s productivity).
Consequently, we include job tenure as a proxy for the level of employee risk aversion,
although we recognize this variable could reflect other factors as well.
Individuals in the data are all full-time employees of a public university in the
southeast during at least one of the years 2001, 2002, and 2003. This university has a
workforce that includes approximately 12,000 full-time employees in a given year, of
which roughly 18 percent are faculty and the rest are considered staff. In each of the
three years, the university offered a FSA to their employees with a contribution minimum
of $250 and a contribution maximum of $4,000. The university also offered a variety of
health plans for employees to choose from. Premiums vary by plan and the university
covers most of the premium for the employee and a smaller percentage of the premium
for any dependents.6 Table 1 gives both the total monthly premium for single coverage
under each plan and the employee’s share.
In 2001, employees could choose between a traditional health maintenance
organization (HMO) and a preferred provider organization (PPO). The HMO network
consists mainly of university providers and involves very little out-of-pocket expenses for
enrollees. The PPO has a wider provider network that includes non-university physicians
and requires more out-of-pocket cost sharing (due to higher coinsurance rates and
deductibles). Table 1 illustrates the fact that the HMO premium is always the lowest
among all plans offered by the university.
In 2002, two new plans were offered, a PPO High option and an exclusive
provider organization (EPO) plan. The PPO High option has a lower deductible than the
PPO and a higher premium. The EPO is similar to the HMO in terms of minimal cost
sharing, but offers a wider provider network. This makes the EPO premium the highest
among all of the plans. In 2003, a consumer-driven plan was offered for the first time.
The consumer-driven plan offers a $500 personal care account, a $1,000 deductible, and
then a 20 percent coinsurance rate that becomes active once the deductible has been met.
The plan also covers preventative care with a $15 co-payment. None of the employees in
our sample in 2003 selected the consumer-driven plan.7
6 For a formal treatment of why an employee’s level of contribution to employer-provided health insurance
premiums is important for the choice of health plan see Gruber and McKnight (2003).
7 In the benefits information provided by the employer, the possibility of rolling over un-used funds from
the personal care account into the next plan year is not mentioned. However, starting in 2004 this is a
possibility, subject to some limits.
Data and Methods
The analysis is focused on the full-time university employees between the ages of
25 and 64 with no missing data that worked at least one year during the three year period
2001-2003. Medicare and other health benefits are available to employees aged 65 and
older. All of the data on these employees was provided by the human resources
department at the university. Table 2 provides the descriptive statistics for the sample.
Table 2 illustrates that the average age among employees in each year is about 45
years old, between 55 percent and 58 percent of employees are female in each year, and
there are slightly more married employees than single. In each year between 15 percent
and 16 percent of employees are non-white and the average job tenure is approximately
11 years. FSA participation in the sample is growing over time, as is the average FSA
contribution.8 In 2003, the average FSA contribution for all employees is $159 dollars.
Among FSA participants, the average contribution in 2003 is $1,257. Figure 1 shows
that most of these contributions are $1,500 or less. The average salary is increasing
between 2001 and 2003.
HMO enrollment is falling slightly over time. The introduction of the PPO High
option and the EPO seemed to draw employees away from the PPO towards these plans
with less cost sharing and higher premiums, but equivalent (broad) provider networks.9
It is also interesting to note that the introduction of these additional insurance options did
not result in a statistically significant change in the percent of employees with no
insurance coverage from the university. Finally, we estimate each employee’s marginal
8 The employer has a separate FSA program for day care expenses. We ignore contributions to the day care
FSA program because in each year less than 3 percent of employees participate. This is much smaller than
the double digit participation rate in the health care FSA program in each year. Therefore, day care
contributions are completely excluded from this analysis.
9 To be more specific, 6.57 percent of 2001 employees switched to the newly introduced PPO High or EPO
options in 2003. Of these plan changers, 20 percent (72 percent) were enrolled in an HMO (PPO) in 2001.
This implies 8 percent of these plan changers were uninsured through this employer in 2001.
tax rate in each year using TAXSIM Version 6.4. The inputs we provided are the tax
year, the employee's salary, their marital status, and their state of residence. The
marginal tax rate estimates include federal, state, and FICA rates. In 2003, the average
marginal tax rate is 37 percent.
As in Cardon and Showalter (2001), we estimate a standard two part model of
FSA participation and contribution levels.11 Assume that the “propensity” of employee i
to participate in the FSA program in year t is given by:
The binary outcome for whether or not employee i participates is given by:
The contribution equation is estimated for those with positive contributions using OLS
with the dependent variable equal to the natural log of the FSA contribution level:
log(_ )|(_ 1).
FSA contributionFSA participationX βε==+
The covariate vector (Xit) consists of those variables described in Table 2.
We first estimate this two part model separately by year to determine whether the
impact of the covariates on FSA participation and contributions remain stable over the
three year period, particularly after the introduction of the new health plan options in
2002. However, employee choices were relatively persistent over time. Approximately 9
percent of employees working full time in 2001 and 2003 changed their FSA contribution
level between those years, while about 7 percent switched to one of the new health plans
in 2002 or 2003. To allow for individual-level correlation over time, we introduce
employee-level random effects θ1i and θ2i:
10 See Feenburg and Coutts (1993) for an introduction to the TAXSIM model.
11 See Duan et al. (1984) for a discussion of the two part model.
FSA contribution FSAparticipationXu
Because there is reason to believe that the FSA participation decision is made
simultaneously with the choice of health plan, we also estimate a McFadden conditional
logit model for each year to predict the joint decision of FSA participation and health
plan choice. This modeling approach has been used extensively in the literature on health
plan choice, with one recent example being Parente et al. (2004).12 In these models,
researchers typically condition health plan choice on the tax adjusted premium each
employee faces for each plan. These tax adjusted premiums are a function of the non-
adjusted premiums that each employee faces and the employee’s individual marginal tax
rate. Therefore, in the conditional logit model, employee marginal tax rates enter through
the tax adjusted premiums rather than directly on the right hand side.
Rather than restricting our attention to health plans alone, we consider the joint
choice of health plan and FSA participation. Assume that, in each year, an employee
chooses between B health benefit “bundles.” Let yib = 1 if employee i chooses bundle b
and yib′ = 0 for b′ ≠ b. In 2001, employees have six choices (B = 6) and in 2002 and 2003
they have eight (B = 8 due to the introduction of new plans):
Bundle 2001 2002 and 2003
1 No FSA, No Health Plan No FSA, No Health Plan
2 No FSA, HMO No FSA, HMO
3 No FSA, PPO No FSA, PPO
12 Other studies include Feldman et al. (1989), Dowd and Feldman (1994 / 1995), and Harris, Schultz, and
4 No FSA, PPO High / EPO
5 FSA, No Health Plan FSA, No Health Plan
6 FSA, HMO FSA, HMO
7 FSA, PPO FSA, PPO
8 FSA, PPO High / EPO
Each employee is assumed to choose the bundle that maximizes their utility, where the
utility of a health “bundle,” Uib is given by:
13 Note that at this employer workers may sign up for the health care FSA even if they are not signed up for
one of the health plans the employer sponsors. Bundle 5 above represents this choice. In may not be the
case in every firm that employees are allowed to sign up for the FSA program without also participating in
an employer-sponsored health plan.
where αb is a bundle-specific constant capturing the average contribution to utility, and
Wi is a vector of employee characteristics, such as gender and race, which do not vary
across bundles. The impact of employee characteristics on utility is allowed to vary
across bundles by allowing for a bundle-specific parameter vector, γb. We can then
estimate the effect of each employee characteristic on the probability of choosing each
Employee choice will also be a function of a vector of bundle-specific personal
attributes, Mib, that vary by bundle and by employee. The two bundle-specific personal
attributes we consider are the bundle’s “tax adjusted” out-of-pocket insurance premium
and the bundle’s FSA tax savings. The bundle’s tax adjusted out-of-pocket premium is
defined to be the premium paid by the employee for the health plan in the bundle,
adjusting for the fact that employees purchase health insurance with pre-tax dollars. The
bundle’s FSA tax savings are defined to be how much tax savings the employee would