The Long-Term Effects of the
Minnesota Family Investment Program
on Marriage and Divorce
Among Two-Parent Families
Prepared by:
Lisa A. Gennetian
Senior Research Associate
MDRC
16 East 34
th
Street
New York, NY 10016
October 2003
Prepared for:
U.S. Department of Health and Human Services
Office of the Assistant Secretary for Planning and Evaluation
200 Independence Avenue, SW
Washington, DC 20201
Contract No. HHS-100-99-0008
Under subcontract to The Lewin Group
-iii-
Table of Contents
List of Tables and Figures iv
Executive Summary ES-1
Introduction 1
The MFIP Model and Evaluation 1
Brief Review of the Final Report’s Key Findings at the 36-Month Follow-Up Point 3
Overview of Data Collection for Long-Term Follow-Up Analyses on
Marriage and Divorce 4
Key Findings on the Effects of MFIP on Marriage and Divorce During a Seven-Year
Follow-Up Period 5
Conclusions 23
Appendix 25
References 33
-iv-
List of Tables and Figures
Table
1 Effects on Marriage and Divorce for Two-Parent Families over a
Seven-Year Follow-Up Period 7
2 Effects on Marriage and Divorce Status for Two-Parent Families at the
End of the Follow-Up Period 9
3 Effects on Divorce for Two-Parent Recipient Families over a
Seven-Year Follow-Up Period, by Subgroup 13
4 Effects on Divorce for Two-Parent Recipient Families Who Were
Married at Baseline, over a Seven-Year Follow-Up Period 16
5 Effects on Marriage and Divorce for Two-Parent Recipient Families
Who Were Cohabiting at Baseline, over a Seven-Year Follow-Up Period 22
Figure
1 Cumulative Percentage Ever Divorced for Two-Parent Recipient Families over a
Seven-Year Follow-Up Period 10
2 Cumulative Percentage Ever Divorced for Two-Parent Recipient Families in Calendar
Months: Comparing Pre-June 1998 and Post-June 1998, When the Pilot
Study Ended and MFIP-S Was Implemented 11
3 Cumulative Percentage Ever Divorced for Two-Parent Recipient Families Who Were
Married at Baseline, over a Seven-Year Follow-Up Period 15
4 Cumulative Percentage Ever Married for Two-Parent Recipient Families
Who Were Cohabiting at Baseline, over a Seven-Year Follow-Up Period 18
5 Marital Status in Each Month of Follow-Up for Two-Parent Recipient Families
Who Were Cohabiting at Baseline 19
6 Cumulative Percentage Ever Divorced for Two-Parent Recipient Families Who Were
Cohabiting at Baseline, over a Seven-Year Follow-Up Period 20
Appendix Table
1 Selected Characteristics of Two-Parent Sample Members,
by Welfare Status at Random Assignment 26
2 Effects on Marriage and Divorce for Two-Parent Recipient Families
over a Seven-Year Follow-Up Period, Year by Year 29
3 MFIP’s Effects on Marriage and Divorce for Two-Parent Applicant Families
over a Six-Year Follow-Up Period, Year by Year 30
4 Effects on Divorce for Two-Parent Applicant Families
over a Six-Year Follow-Up Period 31
ES-1
Executive Summary
In 1994, Minnesota began to test a major welfare reform initiative that emphasized fi-
nancial incentives for work, a participation requirement for long-term recipients, and the simpli-
fication of rules and procedures for receiving public assistance. This program, called the Minne-
sota Family Investment Program (MFIP), was initially implemented in seven counties. MDRC
conducted an in-depth evaluation of MFIP’s effectiveness and impact on various populations
served, using a random assignment design that placed over 14,000 families in either the MFIP
or the Aid to Families with Dependent Children (AFDC) system. The evaluation has produced
findings on participants’ employment, earnings, welfare receipt, income, and other measures of
children’s and parents’ outcomes over a three-year follow-up period for single- and two-parent
families.
1
One of the striking findings of this evaluation was that a survey sample of two-
parent recipient families assigned to MFIP was 19.1 percentage points, or 40 percent, more
likely to be married at the three-year follow-up point than two-parent recipient families as-
signed to AFDC. The three-year follow-up evaluation left open a number of important ques-
tions: Did these increases in marital stability represent short-term delays in separation and di-
vorce, or did they point to lasting effects? What were the effects on marital stability and divorce
among subgroups of two-parent families, especially among those who were cohabiting versus
those who were married at study entry? The specific implications of these findings for families
and children depend on the answers to these questions.
Under subcontract to The Lewin Group, MDRC received funding from the Office of
the Assistant Secretary for Planning and Evaluation, U.S. Department of Health and Human
Services, to examine the effects on divorce and marriage outcomes over a seven-year follow-up
period for the full sample of two-parent families who were part of the MFIP pilot study. Data
for the analyses come from publicly available divorce and marriage certificate records.
1
Miller et al., 2000.
ES-2
Key Findings
Two-parent recipient families are defined as families who were receiving or reapplying
for welfare, and were either married, or living together with a shared biological child, when they
entered the MFIP study. Among two-parent recipient families:
• MFIP decreased divorce by 3.5 percentage points, or by about 25 percent,
seven years after study entry. Effects on divorce primarily occurred late
in the follow-up period, even after June 1998, when the MFIP pilot study
ended and a statewide MFIP program was implemented.
• For two-parent recipients who were married at study entry, MFIP in-
creased marital stability by decreasing divorce.
• Among cohabiting couples — coupled parents who shared a biological
child at study entry — the cumulative rate of ever marrying during the
seven-year follow-up period was similar for the MFIP and AFDC (con-
trol) groups.
• However, MFIP cohabiting couples were 66 percent less likely than
AFDC cohabiting couples to get divorced during the follow-up period.
Thus, the proportion of cohabiting couples that were married at the end
of the follow-up was higher among MFIP families than among AFDC
families.
• MFIP’s effects on marital stability were most pronounced among black
recipient families, reducing rates of divorce among black married couples
by over 70 percent. Effects on marital stability did not vary by family size,
prior marital status, prior employment history, or prior welfare history of
the family.
Two-parent applicants are defined as two-parent families who were newly applying for
welfare at the time they entered the MFIP study. Among two-parent applicant families:
• Within the AFDC (control) group, two-parent applicant families were
more likely to divorce, and less likely to marry, than two-parent recipi-
ent families. Over the seven-year follow-up period, rates of divorce
across two-parent applicant families in the control group, at about 19
percent, were five percentage points higher than rates of divorce among
recipient families. Among cohabiters, rates of marriage for applicant
ES-3
couples in the control group, at 17 percent, were four percentage points
lower than rates of marriage among recipient couples.
• MFIP had no cumulative effect on divorce among two-parent applicant
families, but did somewhat increase divorce late in the follow-up period.
In the original MFIP evaluation, the effects of the program on marital stability were
measured for 290 two-parent recipient families who were respondents to a 36-month follow-up
survey. The new analysis presented here has expanded on these findings by providing long-term
follow-up for the full sample of 1,515 two-parent recipient families and 731 two-parent appli-
cant families who participated in the evaluation. Overall, the results indicate that the pilot MFIP
program that began in 1994 continued to have effects on rates of divorce for two-parent families
seven years after they entered the study. Reductions in divorce were concentrated among fami-
lies who were already receiving welfare when they entered the study rather than new welfare
applicants. Impacts were most pronounced for black parents who were already married at the
time they entered the study, and for parents who were cohabiting when they entered the study.
Note that because this new analysis relies solely on public records of marriages and fi-
nalized divorces, it does not capture any effects that the program may have had on couples’ like-
lihood of separating or living apart without formally divorcing. Nevertheless, these findings
represent some of the best evidence to date about the potential for welfare policies to affect
marital stability among two-parent families.
The results raise several important questions for future work. First, because most wel-
fare reform evaluations in the 1990s did not collect information on two-parent families, there is
little impact information available for two-parent families, making it important to replicate this
type of evaluation in some additional geographic areas to determine whether these findings are
generalizable beyond Minnesota. Second, to further understand the results presented here, future
work will examine how MFIP affected the timing of marriages or divorces, particularly for co-
habiting parents. In addition, the suggestion of some increase in divorce in some years, among
applicant families, is worthy of further investigation.
Finally, the effects on divorce among MFIP’s two-parent recipient families suggest that
the program could have had important effects on the well-being of children in these families.
With funding from various foundations, via administrative records (including child welfare re-
cords), MDRC is continuing to conduct long-term analyses on the effects of the pilot MFIP
program on economic, family, and child outcomes among single-parent and two-parent fami-
lies, with the expectation that additional findings will be available in 2004.
-1-
Introduction
In 1994, Minnesota began a major welfare reform initiative that emphasized financial
incentives for work, a participation requirement for long-term recipients, and the simplification
of rules and procedures for receiving public assistance. This program, called the Minnesota
Family Investment Program (MFIP), was initially implemented in seven counties. MDRC con-
ducted an in-depth evaluation of MFIP’s effectiveness and impact on various populations
served, using a random assignment design that placed over 14,000 families in either the MFIP
or the Aid to Families with Dependent Children (AFDC) system. The evaluation has produced
findings on participants’ employment, earnings, welfare receipt, income, and other measures of
children’s and parents’ outcomes over a three-year follow-up period for single- and two-parent
recipient families.
2
One of the striking findings of this evaluation was that a survey sample of
two-parent recipient families in MFIP was 19.1 percentage points, or 40 percent, more likely
than two-parent recipient families in AFDC to be married at the three-year follow-up point.
Under subcontract to The Lewin Group, MDRC received funding from the Office of
the Assistant Secretary for Planning and Evaluation, U.S. Department of Health and Human
Services, to examine the effects on marriage and divorce outcomes over a seven-year follow-up
period — using data from publicly available divorce and marriage certificate records — for the
full sample of two-parent families who were part of the MFIP pilot program. These two-parent
families included those who were married at study entry as well as those who cohabited and
shared a biological child.
The MFIP Model and Evaluation
MFIP integrated several existing programs in the Minnesota welfare system.
3
These in-
cluded not only AFDC (the core of the traditional system), but also STRIDE, the state’s em-
ployment and training program for AFDC recipients (which operated on a voluntary basis for
certain targeted groups); the Family General Assistance (FGA) program, a state-run program
which allowed some low-income families not eligible for AFDC to qualify for welfare; and the
federally funded Food Stamp Program, which provided assistance in the form of food coupons.
MFIP differed from the AFDC system in three fundamental ways: (1) it decreased the extent to
which families’ welfare grants were reduced when they went to work, thereby making work pay
more effectively; (2) it required two-parent families who had received assistance for six months
in a year to participate in employment and training activities in order to continue receiving their
2
Miller et al., 2000.
3
Miller et al., 2000.
-2-
full grants (though, because the job search/Community Work Experience Program was manda-
tory, the introduction of MFIP employment and training requirements was not a dramatic
change for two-parent families); and (3) it simplified program rules by combining the benefits
of AFDC, FGA, and the Food Stamp Program into a single program, giving food stamp bene-
fits as part of the cash grant, and removing any work history requirements or work effort limi-
tations that existed under the AFDC-Unemployed Parent (AFDC-UP) program.
For purposes of the evaluation, two-parent families were defined as those in which mar-
ried or cohabiting parents (either biological parents or stepparents of at least one child in the
family) were living in the home at the time of random assignment. These two-parent families
were randomly assigned into either the MFIP group or the AFDC group at the time of recertifi-
cation or application for receipt of welfare benefits. In each case, the second parent was defined
by the Social Security number provided by the first parent when baseline forms were submitted
at the time of recertification or application for receipt of welfare benefits.
All two-parent families assigned to the MFIP group received MFIP benefits, which, in
addition to providing financial incentives similar to those for single-parent families, removed
significant restrictions on eligibility in the AFDC-UP program, including the work history re-
quirement and the 100-hour rule.
4
When these families had received public assistance for 6 of
the previous 12 months, at least one parent was required to participate in MFIP’s employment
and training services. Two-parent families in the AFDC group were eligible for the benefits and
services of the AFDC system (primarily AFDC-UP)
5
and the Food Stamp Program as de-
scribed above.
Appendix Table 1 provides basic demographic information about the two-parent fami-
lies who were members of the study sample. The majority of two-parent families in the MFIP
pilot resided in urban counties in and around Minneapolis. Two-thirds of recipient families had
received welfare for two years or more at the time they entered the study — a longer history on
welfare than was true of the national caseload in 1995, of which less than 40 percent of two-
4
In most two-parent families, both biological parents were present, and the family was evaluated for eligibil-
ity for AFDC-UP. To be eligible for AFDC-UP, the family had to document that the primary wage earner had
worked in at least 6 of the previous 13 calendar quarters (the “work history” requirement) and had been unem-
ployed for at least 30 days prior to approval for benefits. In addition to these restrictions, the two-parent family had
to be financially eligible for benefits; if the primary wage earner worked while receiving AFDC-UP benefits, he or
she was limited to working no more than 100 hours per month (the 100-hour rule).
5
Families were also eligible for the AFDC-Incapacitated Program (AFDC-INCAP), if one parent was inca-
pacitated. A small proportion of two-parent families in the AFDC group received cash assistance from the FGA
program instead of from AFDC.
-3-
parent families had been continuously on welfare for two years or more.
6
Characteristics of re-
cipients also differed from those of applicants in the MFIP sample. For example, nearly 80 per-
cent of the two-parent applicant families are white, compared with 60 percent of two-parent re-
cipient families, and close to 80 percent of applicants were married at study entry, versus 69
percent of two-parent recipient families.
Brief Review of the Final Report’s Key Findings at the 36-Month
Follow-Up Point
The final report of the MFIP evaluation presents and summarizes MFIP’s effects at the
36-month follow-up point on employment, income, marriage, and other family outcomes, for
two-parent recipient and applicant families.
7
All noted effects are estimated by comparing re-
gression-adjusted outcomes between families who were randomly assigned, in a lottery-like
process, to either the MFIP group or the AFDC group.
8
In brief, the key findings from the 36-month follow-up were:
• Two-parent recipient families in MFIP were as likely as those in the AFDC
group to have at least one parent work, but less likely to have both parents
work, leading to lower combined earnings.
• MFIP two-parent recipient families were 40 percent more likely to be mar-
ried and living together than their counterparts in the AFDC group.
• Compared with AFDC recipient families, MFIP two-parent recipient families
had higher incomes, and twice as many of them owned a home at the end of
the study’s third year.
6
Note, however, that the length of stay on welfare for recipients in the research sample is partly an artifact of
the way random assignment was conducted. Because random assignment of recipients took place at annual recer-
tification interviews, most recipients, by definition, should have been on welfare for at least one year at baseline.
7
Miller et al., 2000; Knox et al., 2000.
8
All effects or impacts were estimated using ordinary least squares techniques and regression-adjusted, for
precision, with the following baseline and pre-random assignment covariates: county of residence, recipient status,
marital status, number of children, receipt of high school degree or General Educational Development (GED)
certificate, race/ethnicity, age, quarter of random assignment, current employment of primary parent, prior welfare
receipt, and prior employment of each parent. Prior work suggests that impacts on marriage outcomes estimated
using other nonlinear techniques, such as probit or logit techniques, are similar (e.g., see Harknett and Gennetian,
in press). Effects were tested for statistical significance using a two-tailed t-test. Unless otherwise noted, only sta-
tistically significant effects are discussed.
-4-
•
Two-parent applicant families in MFIP were as likely to have at least one
parent work, but less likely to have both parents work, leading to lower com-
bined earnings.
9
The effects of MFIP on marriage from the 36-month follow-up were estimated for a
survey sample of 290 two-parent recipient families and were based on the survey measure “In
the prior month, were you married and living with a spouse?” In addition, effects on marriage
were confirmed via manually collected, publicly available divorce records data for the subset of
the survey sample married at study entry. The survey sample of two-parent applicant families
was too small to evaluate program impacts on marriage or other family outcomes.
Overview of Data Collection for Long-Term Follow-Up Analyses
on Marriage and Divorce
The data collection for the long-term follow-up analyses occurred in two phases. First,
in early September 2001, MDRC obtained, via electronic files, publicly available divorce re-
cords — names and dates of divorces that took place in the state of Minnesota from January
1994 until August 2001 — from the Minnesota Supreme Court.
10
Second, in March 2003, upon
completion of a contract with the Minnesota Department of Health, MDRC obtained, via elec-
tronic files, certificate records of marriages that took place in the state of Minnesota from Janu-
ary 1989 until December 2001. Approximately 96 percent of MFIP two-parent family recipients
were randomly assigned from April 1994 to March 1995, and thus, analyses with these two data
sources cover a roughly 6.5-year to 7.5-year follow-up period. The average follow-up period is
shorter among MFIP two-parent family applicants, because two-parent family applicants were
randomly assigned from April 1994 to March 1996, with nearly 30 percent being randomly as-
signed from April 1995 to March 1996.
The marriage and divorce records data were checked, cleaned, and substantially refor-
matted by MDRC’s information specialists in order to match the names of each spouse in the
marriage or divorce records data to the names of each spouse or partner in the pilot MFIP two-
parent family sample. Names of respondents and their spouses in the pilot MFIP evaluation, and
any variation in the spelling of these names, were obtained from Background Information
Forms (BIFs) and Unemployment Insurance earnings records (from Minnesota’s Department of
9
As noted in Miller et al., 2000, the survey sample of two-parent applicant families was too small to analyze
other family outcomes at the 36-month follow-up point, such as material hardship, health insurance coverage,
home ownership, and marital stability.
10
One key benefit of obtaining the divorce records data electronically is that all resulting analyses capture any
divorce that occurred in any county in the state rather than the seven counties that participated in the pilot MFIP
evaluation. The availability of state data eliminates any bias that may have been generated by intra-state moves by
sample members during the follow-up time period.
-5-
Economic Security). The first and last name of each spouse in an MFIP two-parent family was
then matched to the first and last name of each spouse in the marriage and divorce records data.
In addition, information about the birth date and Social Security number of the bride — when
available and valid — was used to confirm matches in the marriage records analysis.
Reports about marital status in the BIF (completed by staff in the welfare offices via
client interview just prior to random assignment) and successful matches with the marriage cer-
tificate records were used to construct the marriage records file.
11,12
Both sources were relied
upon because many marriages may have taken place prior to the time period in which marriage
certificates data were available for analysis, i.e., before 1989. After numerous quality checks on
the data and on the matches, a total of 329 finalized divorces were found from April 1994 to
August 2001 for the entire two-parent family sample (2,246 two-parent families) in the MFIP
pilot evaluation, including all program and control group families. This translates to an overall
divorce rate of approximately 15 percent over a roughly seven-year period for this sample. Ap-
proximately 195 finalized divorces occurred within the sample of 1,515 two-parent recipient
families, for an overall divorce rate of about 13 percent, and 134 finalized divorces oc-
curred within the sample of 731 two-parent applicant families, for an overall divorce rate
of 18 percent.
Key Findings on the Effects of MFIP on Marriage and Divorce
During a Seven-Year Follow-Up Period
The effects of MFIP on marriage and divorce during a seven-year follow-up period, us-
ing marriage certificate records data and publicly available divorce records data, are presented
in Tables 1 to 5 and Figures 1 to 6.
13
Effects of MFIP on marriage and divorce, as measured by
11
Note that each sample family’s marital status was checked for consistency using information from divorce
and marriage records, and the BIF. With these three sources of information, the following coding decisions were
implemented: (1) ten sample members with missing baseline information about marital status are excluded from
the analysis; (2) three sample members who had two dates of marriage documented in the marriage certificates
data were coded as married at the first noted date, unless a divorce was also documented; (3) four couples who
were recorded on the BIF as cohabiting were recoded as married at baseline. In addition, one family who was
recorded on the BIF as married, and for whom a divorce was recorded with divorce records data at around the
time of random assignment, was coded as divorced one month after random assignment. The findings do not
change if we assume that this divorce occurred at a different time (i.e., at or much later than the time of random
assignment).
12
No record of marriage was found for 13 cohabiting two-parent families who were divorced according to the
match with the divorce records data. This may be because the marriages took place out of the state of Minnesota
or because the quality of information was not adequate to secure a match with the Minnesota marriage records.
13
Note that these findings are not based on the same measure as the findings reported from the 36-month fol-
low-up: Prior findings were based on a survey respondent’s stating that she and her partner were “married and
living together,” while the current findings are based on data from marriage certificates and divorce records.
-6-
differences between families in the program group and families in the control group, are shown
for the total sample of two-parent recipient families and two-parent applicant families. Effects
are presented according to marital status at the time of study entry for recipient families and
then applicant families, and by various other subgroups according to characteristics of the fami-
lies at the time of study entry; those characteristics include race/ethnicity, number of children in
the family, age of youngest child in the family, and prior welfare and employment history.
Note that for those who were married at baseline, the primary outcome examined is the
program’s impact on their likelihood of divorcing during the follow-up period. However, for
those who were cohabiting at baseline, we are interested in both the likelihood of marriage and
the likelihood of subsequent divorce. We present the program’s impacts on marriage from two
different perspectives. One is to estimate the couple’s likelihood of ever marrying during the
follow-up period, a proportion that cumulates over time as more marriages are observed. The
second is to estimate the couple’s likelihood of being married at a given point in time, particu-
larly the end of the follow-up period. When estimating the percentage who are married at a
point in time, couples are not counted as “married” in the time period after a divorce has been
observed. Thus, the program’s impact on cohabiters’ likelihood of ever marrying may differ
from its impact on the percentage who are married at the end of the follow-up period.
Effects on two-parent recipient families overall
• MFIP decreased divorce by 3.5 percentage points, or by about 25 percent,
seven years after study entry, indicating a lasting reduction in divorce.
Table 1 shows that 13.8 percent of AFDC recipient families had a finalized divorce by
the seven-year follow-up point, compared with 10.4 percent of MFIP recipient families, a 3.5
percentage point difference, or a 25 percent reduction. These findings confirm that MFIP’s ef-
fects on divorce held up for the full sample of two-parent recipient families. That is, effects
were not limited to the 36-month follow-up survey sample of 290 families.
14
In addition, the
effects have lasted well beyond the time period covered in the final evaluation report. Note that
because virtually no couples who divorced were remarried to one another by the end of the fol-
14
Long-term follow-up results for the sample of 290 two-parent recipient families who were surveyed at the
36-month follow-up point (see Miller et al., 2000, for a more detailed description of this survey sample) are as
follows: Approximately 18.2 percent of AFDC recipient families had a finalized divorce at the seven-year follow-
up point, compared with 12.4 percent of MFIP recipient families, for a 5.8 percentage point difference, or 46.8
percent reduction. This effect is not statistically significant at the seven-year follow-up point (p=0.20), very possi-
bly due to the small sample size. Note also that the general magnitude of this effect is quite similar to what was
reported for this survey sample at the five-year follow-up point (see Table 6.6, page 173, in Miller et al., 2000).
-7-
Sample Percentage Percentage
Size MFIP AFDC Impact
a
Change MFIP AFDC Impact
a
Change
Recipient families
b
1,515 10.4 13.8 -3.5 ** -25.0
Married at baseline 1,043 13.8 16.8 -2.9 -17.5
Cohabiting at baseline 472 2.7 8.0 -5.3 ** -66.2 20.6 20.8 -0.3 -1.3
Applicant families
c
731 21.0 18.8 2.2 11.7
Married at baseline 577 25.0 23.5 1.5 6.3
Cohabiting at baseline 154 6.7 3.5 3.1 89.9 23.4 16.8 6.6 39.1
MFIP's Long-Term Effects on Marriage and Divorce
{ Not applicable }
{ Not applicable }
Table 1
Ever Married (%)Ever Divorced (%)
Effects on Marria
g
e and Divorce for Two-Parent Families over a Seven-Year Follow-Up Period
SOURCES: Calculations using public divorce and marriage certificate records from the State of Minnesota.
N
OTES: The sample includes members randomly assigned from April 1, 1994 to March 31, 1996, excluding the small percentage who
were receiving or applying only for food stamps at random assignment. Approximately 96 percent of two-parent recipient families were
randomly assigned from April 1994 to March 1995.
A two-tailed t-test was applied to regression-adjusted impact estimates. Statistical significance levels are indicated as
***= 1 percent; **= 5 percent; *= 10 percent.
Rounding may cause slight discrepancies in sums and differences.
a
The difference is the impact of the financial incentives, mandatory services, reinforced incentive messages, and elimination of the
100-hour rule and work history requirement.
b
Recipients have an average of 7.3 years of follow-up for marriage and an average of 6.9 years for divorce.
c
Applicants have an average of 7.0 years of follow-up for marriage and an average of 6.7 years for divorce.
-8-
low-up period, any differences between the impacts on divorce in Tables 1 and 2 are caused by
a drop in the sample that had the full years of follow-up required for the point-in-time estimates
presented in Table 2.
• MFIP’s effects on divorce primarily occurred late in the follow-up period,
even after June 1998, when the MFIP pilot study ended and a statewide
MFIP program was in place.
Figure 1 shows that rates of finalized divorces in MFIP families were lower from month
26 to month 50, and from month 60 to month 83, compared with the same rates for AFDC fami-
lies, with these differences being significant in the sixth and seventh year of the follow-up (see
Appendix Table 2). Figure 2 further shows that reductions in the rates of finalized divorces pri-
marily occurred after June 1998, when the MFIP pilot ended and a statewide MFIP program
was in place.
15
After June 1998, all two-parent families in the pilot MFIP evaluation were sub-
ject to the benefits and requirements of the new statewide MFIP program.
16
Given that the MFIP pilot program ended in 1998, how could it have such prolonged ef-
fects? Although few previous evaluations have found impacts on marriage, one might expect
that MFIP and AFDC families would look similar four years after the end of the pilot program;
an analogy to this situation would be the common occurrence of control group catch-up in em-
ployment impacts, in which a control group experiences the same effects as a program group
after the program group no longer receives a given intervention. The complication in this exam-
ple is that both groups in the pilot program were subjected to the benefits and requirements of a
new program, MFIP-S (Minnesota’s current welfare reform program).
There are a number of possible explanations for the persistence of the impacts. First, the
pilot program (with its streamlined eligibility rules, generous earnings disregard, and consolida-
tion and cashing out of welfare benefits) could have prompted a permanent change in two-
parent recipient families that set them on a different path. Second, statewide MFIP — with rela-
tively less generous benefits than the original MFIP — may not have provided an environment
that allowed AFDC families to stabilize or increase marriage rates. Third, even if statewide
MFIP has the potential to affect two-parent families positively, it may be that timing matters,
and that it was too late for the marriages of AFDC families in the pilot program to be signifi-
cantly affected by the changes instigated by the statewide program. Fourth, many of the two-
15
MFIP-S, the statewide program, was implemented in January 1998. However, MFIP pilot study members
were exempt from the statewide program until June 1998. For this analysis the data were converted back into cal-
endar months. Thus, the “relative” length of follow-up differs slightly for each two-parent family because most of
the two-parent recipient families were randomly assigned over a 12-month period.
16
Some aspects of MFIP-S that applied to MFIP pilot families were implemented in July 1997. For example,
the elimination of the 100-hour rule applied to all MFIP pilot and AFDC families starting in July 1997.
-9-
Sample Percentage Percentage
Size MFIP AFDC Impact
a
Change MFIP AFDC Impact
a
Change
Recipient families 1,498 8.7 11.7 -3.0 ** -29.5
Married at baseline 1,038 11.7 15.0 -3.3 -21.7
Cohabiting at baseline 467 1.7 5.5 -3.8 ** -68.9 18.6 17.3 1.4 7.8
Applicant families 731 16.2 11.9 4.3 * 35.8
Married at baseline 577 19.8 14.2 5.6 * 39.1
Cohabiting at baseline 154 4.4 4.0 0.4 10.4 20.1 13.8 6.3 46.0
MFIP's Long-Term Effects on Marriage and Divorce
{ Not applicable }
{ Not applicable }
Table 2
Effects on Marriage and Divorce Status for Two-Parent Families at the End of the Follow-Up Period
Divorced at End of Year 6 (%) Married at End of Year 6 (%)
Divorced at End of Year 5 (%)
Married at End of Year 5 (%)
SOURCES: MDRC Calculations using public divorce and marriage certificate records from the State of Minnesota.
N
OTES: The sample includes members randomly assigned from April 1, 1994 to March 31, 1996, excluding the small percentage who were
receiving or applying only for food stamps at random assignment. Approximately 96 percent of two-parent recipient families were randomly
assigned from April 1994 to March 1995.
A two-tailed t-test was applied to regression-adjusted impact estimates. Statistical significance levels are indicated as
***= 1 percent; **= 5 percent; *= 10 percent.
Rounding may cause slight discrepancies in sums and differences.
Because of the long intake period, this table excludes recipients who did not have data for six full years and applicants who did not hav
e
data for six full years.
a
The difference is the impact of the financial incentives, mandatory services, reinforced incentive messages, and elimination of the 100-
hour rule and work history requirement.
-9- -10-
MFIP's Long-Term Effects on Marriage and Divorce
Cumulative Percentage Ever Divorced for Two-Parent Recipient Families
Figure 1
over a Seven-Year Follow-Up Period
-5
0
5
10
15
20
Months After Random Assignment
Percentage
MFIP
AFDC
Impact
SOURCES: MDRC calculations using public divorce and marriage certificate records from the State of Minnesota.
N
OTE: The sample includes members randomly assigned from April 1, 1994 to March 31, 1996, excluding the small
percentage who were receiving or applying only for food stamps at random assignment. Approximately 96 percent of two-
parent recipient families were randomly assigned from April 1994 to March 1995.
2 6 12 18 24 30 36 42 48 54 60 66 72 78 84
-11-
MFIP's Long-Term Effects on Marriage and Divorce
When the Pilot Study Ended and MFIP-S Was Implemented
Figure 2
Cumulative Percentage Ever Divorced for Two-Parent Recipient Families in Calendar Months:
Comparing Pre-June 1998 with Post-June 1998,
-10
-5
0
5
10
15
20
Percentage
Impact
MFIP
Jul '94 Jun '95 Apr '96 Mar '97 Feb '98 Jan '99 Dec '99 Nov '00 Oct '01 Dec '01
AFDC
SOURCES: MDRC calculations using public divorce and marriage certificate records from the State of Minnesota.
N
OTE: The sample includes members randomly assigned from April 1, 1994 to March 31, 1996, excluding the small
percentage who were receiving or applying only for food stamps at random assignment. Approximately 96 percent of two-
parent recipient families were randomly assigned from April 1994 to March 1995.
June 1998
-12-
parent recipient families may have left welfare after four years or more, limiting the possibility
for the statewide program to have any effects. Some of these possibilities will be investigated in
future work examining the long-term economic impacts of the program.
It is likely that the persistence of effects was caused by some combination of all of these
factors. Furthermore, because one would expect some control group catch-up to occur after June
1998, it may be that the current estimates on divorce are a lower bound of what may have hap-
pened had statewide MFIP not been implemented: Movement of the control group into the
statewide MFIP program may have decreased the differences that would have occurred had the
treatment difference continued indefinitely.
• MFIP particularly reduced rates of divorce among families whose young-
est child, at study entry, was less than 6 years old.
Table 3 presents MFIP’s effects on divorce for two-parent recipient families over a
seven-year follow-up, for various subgroups. The first panel of Table 3 shows the effects of
MFIP on divorce by the number of children in the recipient family. One hypothesis is that if
MFIP’s streamlined eligibility rules (e.g., the elimination of the 100-hour rule) were an impor-
tant policy component driving MFIP’s effects on marital behavior, we should see larger reduc-
tions in divorce for large families than small families. Under the 100-hour rule, large families —
who may have found it difficult to make enough income working part time in a low-wage job
— have a larger incentive than small families to divorce in order to maintain welfare eligibility
while working more than 100 hours per month. MFIP’s effects on divorce were slightly more
pronounced for families who had three or more children at study entry, reducing divorce from
12.3 percent for AFDC families to 7.0 percent for MFIP families. However, effects on divorce
for families with three or more children were not significantly different from effects on divorce
for families with fewer than three children, weakening the evidence that streamlined eligibility
rules were primarily responsible for MFIP’s impacts.
17
The second panel of Table 3 shows that recipient families in MFIP who had a young
child at study entry were less likely to divorce, compared with their AFDC counterparts, by 5.0
percentage points, or 35 percent. In contrast, MFIP had no effect on divorce for families who, at
study entry, had a child aged 6 or older.
• MFIP reduced rates of divorce among black recipient families by 59 per-
cent by the end of the seven-year follow-up period.
17
A similar conclusion was drawn based on the 36-month survey data on 290 recipient families (see Chapter
6, Miller et al., 2000).
-13-
P-Value fo
r
Sample Percentage Subgroup
Size MFIP AFDC Impac
t
a
Change Differences
All recipient families 1,515 10.4 13.8 -3.5 ** -25.0
N
umber of children 0.52
Fewer than 3 childre
n
789 12.4 15.5 -3.1 -20.0
3 or more children 699 7.0 12.3 -5.2 ** -42.8
Age of youngest child 0.10
Less than 6 years old 1,152 9.4 14.4 -5.0 *** -34.7
6 years old or older 336 13.2 11.2 2.0 17.8
Race/Ethnicit
y
0.10
White, non-Hispanic 895 14.0 16.1 -2.1 -13.1
Black, non-Hispanic 243 6.8 16.7 -9.9 ** -59.4
Asian/Pacific Islander 239 6.1 5.1 1.0 19.7
Other
b
125 3.7 15.4 -11.7 * -76.1
Employment 1 year prior
to study entr
y
0.84
One parent employed 544 11.2 14.5 -3.4 -23.1
Both parents employed 450 14.5 20.3 -5.9 -28.9
N
o parent employed 521 5.5 9.0 -3.5 -39.3
Welfare receipt prior to
study entr
y
0.41
Less than 2 years 518 13.0 14.4 -1.4 -9.9
2 years to 5 years 455 7.6 14.6 -7.0 ** -48.0
More than 5 years 526 10.1 13.3 -3.1 -23.7
Ever Divorced (%)
Effects on Divorce for Two-Parent Recipient Families
over a Seven-Year Follow-Up Period, by Subgroup
Table 3
SOURCES: Calculations using public divorce and marriage certificate records from the State of Minnesota.
N
OTES: The sample includes members randomly assigned from April 1, 1994 to March 31, 1996, excluding
the small percentage who were receiving or applying only for food stamps at random assignment.
Approximately 96 percent of two-parent recipient families were randomly assigned from April 1994 to March
1995.
A two-tailed t-test was applied to regression-adjusted impact estimates. Statistical significance levels are
indicated as ***= 1 percent; **= 5 percent; * =10 percent.
Rounding may cause slight discrepancies in sums and differences.
Information at baseline on some subgroup characteristics was missing for some sample members.
Therefore, the average impact across subgroups does not always replicate the impact for all recipients.
a
The difference is the impact of the financial incentives, mandatory services, reinforced incentive
messages, and elimination of the 100-hour rule and work history requirement.
b
Hispanic, American Indian, and Alaskan.
-14-
Nearly 60 percent of recipient families identify themselves as white, non-Hispanic; 16
percent as black, non-Hispanic; 16 percent as Asian/Pacific Islander; and about 10 percent as other
ethnicities. Table 3 shows that 16.7 percent of black AFDC families had a finalized divorce com-
pared with 6.8 percent of black MFIP families over the seven-year follow-up period, a 9.9 per-
centage point, or 59 percent, reduction. Though the sample is small, rates of divorce were also
significantly reduced for the combined group of Hispanic, American Indian, and Alaskan families;
in this group, 15.4 percent of AFDC recipient families divorced, compared with 3.7 percent of
MFIP recipient families. There were no impacts on rates of finalized divorces among white, non-
Hispanic and Asian/Pacific Islander AFDC and MFIP recipient families.
Effects on married two-parent recipient families
• MFIP somewhat increased marital stability, by decreasing divorce among
two-parent recipients who were married at study entry.
Table 1 shows that approximately 17 percent of AFDC families who were married at
study entry were divorced by the seven-year follow-up point, compared with 14 percent of
MFIP families. Figure 3 shows that rates of finalized divorce during the seven-year follow-up
period were generally lower for MFIP families than for AFDC families, with significant reduc-
tions occurring during the seventh year (see Appendix Table 2).
• MFIP’s effects on marital stability were most pronounced among black
recipient families, reducing rates of divorce among black married couples
by over 70 percent. Effects on marital stability did not vary by family size,
prior marital status, prior employment history, or prior welfare history of
the family.
Table 4 expands on the analyses presented in Table 3, examining MFIP’s effects on fi-
nalized divorces over a seven-year follow-up period across subgroups (those with adequate
sample sizes), for two-parent recipients who were married at study entry. The variation in di-
vorce rates for members of the control group provides helpful context for interpreting the pro-
gram’s impacts on divorce for these subgroups. In particular, within the control group, the rate
of divorce over the seven-year follow-up was substantially higher for families who were black,
non-Hispanic compared with whites or Asian/Pacific Islanders. MFIP particularly increased
marital stability for this group. Table 4 shows that 28.2 percent of black married AFDC families
ever divorced during the follow-up period, compared with 7.8 percent of black married MFIP
families, for a 21 percentage point, or 73 percent, reduction. MFIP’s effects on divorce for mar-
ried recipients did not vary by family size or by the prior marital, employment, or welfare his-
tory of the family.
-15-
MFIP's Long-Term Effects on Marriage and Divorce
Cumulative Percentage Ever Divorced for Two-Parent Recipient Families Who Were Married
at Baseline, over a Seven-Year Follow-Up Period
Figure 3
-10
-5
0
5
10
15
20
Months After Random Assignment
Percentage
AFDC
MFIP
Impact
SOURCES: MDRC calculations using public divorce and marriage certificate records from the State of Minnesota.
N
OTE: The sample excludes the small percentage who were receiving or applying only for food stamps at random
assignment. Approximately 96 percent of two-parent recipient families were randomly assigned from April 1994 to March
1995.
2 6 12 18 24 30 36 42 48 54 60 66 72 78 84
-16-
P-Value fo
r
Sample Percentage Subgroup
Size MFIP AFDC Impac
t
a
Change Differences
All recipient families 1,043 13.8 16.8 -2.9 -17.5
N
umber of children 0.6
Fewer than 3 children 519 17.8 20.0 -2.2 -11.1
3 or more children 504 9.0 13.3 -4.4 -32.6
Marital history prior to study entry
b
0.6
Married within 5 years 189 29.9 29.1 0.8 2.7
Married 5 years or more 854 10.7 13.8 -3.1 -22.6
Race/Ethnicit
y
0.03 **
White, non-Hispanic 587 20.7 19.3 1.3 6.8
Black, non-Hispanic 155 7.8 28.2 -20.5 *** -72.5
Asian/Pacific Islander 229 5.4 5.1 0.3 6.2
Age of youngest child 0.2 *
Less than 6 years old 724 13.7 18.1 -4.5 -24.7
6 years old or older 299 14.1 12.1 2.0 16.6
Employment 1 year prio
r
to study entr
y
0.4
One parent employed 348 17.8 15.8 2.0 12.8
Both parents employed 292 20.9 26.8 -6.0 -22.2
N
o parent employed 403 7.1 10.5 -3.3 -31.8
Welfare receipt prior to study entr
y
0.4
Less than 2 years 365 15.9 17.3 -1.4 -8.2
2 years to 5 years 313 11.2 19.0 -7.8 * -41.1
More than 5 years 354 14.4 14.7 -0.3 -2.2
MFIP'
s
L
on
g
-
T
erm
Eff
ects on
M
arr
i
a
g
e an
d
Di
vorce
Ever Divorced (%)
Eff
ects on
Di
vorce
f
or
T
wo-
P
arent
R
ec
i
p
i
ent
F
am
ili
es
Wh
o
W
ere
M
arr
i
e
d
at
B
ase
li
ne,
over a
S
even-
Y
ear
F
o
ll
ow-
U
p
P
er
i
o
d
Table 4
SOURCES: MDRC calculations using public divorce and marriage certificate records from the State of
Minnesota.
N
OTES: The sample includes members randomly assigned from April 1, 1994 to March 31, 1996, excluding the
small percentage who were receiving or applying only for food stamps at random assignment. Approximately 96
percent of two-parent recipient families were randomly assigned from April 1994 to March 1995.
A two-tailed t-test was applied to regression-adjusted impact estimates. Statistical significance levels are
indicated as ***= 1 percent; **= 5 percent; * =10 percent.
Rounding may cause slight discrepancies in sums and differences.
Information at baseline on some subgroup characteristics was missing for some sample members.
Therefore, the average impact across subgroups does not always replicate the impact for all recipients.
Subgroups with fewer than 150 families are not reported.
a
The difference is the impact of the financial incentives, mandatory services, reinforced incentive
messages, and elimination of the 100-hour rule and work history requirement.
b
Marital history is determined based on matches with the marriage certificate records and thus may not
precisely capture the timing of every marriage as reported on the Background Information Form.
-17-
Effects on cohabiting recipient families
• For MFIP cohabiting couples — coupled parents who shared a biological
child at study entry — the proportion who ever married during the fol-
low-up period was similar to that for AFDC cohabiting couples.
Table 1 and Figure 4 show that cumulative rates of marriage (i.e., the percentage who
ever married, without adjusting for subsequent divorce) among MFIP cohabiting couples and
AFDC cohabiting couples were similar through the seven-year follow-up period.
• However, MFIP cohabiting couples were 66 percent less likely than
AFDC cohabiting couples to divorce at some point during the follow-up
period. As a result, the proportion of cohabiting couples that were mar-
ried at the end of the follow-up was higher among MFIP families than
among AFDC families.
As shown in Table 1, approximately 8 percent of AFDC cohabiting families had a final-
ized divorce by the seven-year follow-up point, compared with 2.7 percent of MFIP cohabiting
families, for a 5.3 percentage point, or 66 percent, reduction in divorce. Beginning about two years
after random assignment, rates of finalized divorces were consistently higher through most of the
follow-up period for AFDC cohabiting families, compared with MFIP cohabiting families (see
Figure 6). Consequently, as shown in Figure 5, the proportion of cohabiting couples that were
married at the end of the follow-up was slightly higher among MFIP families than among AFDC
families. The ultimate effect of MFIP on marriage among cohabiting couples — an increase in
marital stability — is similar to MFIP’s effect for married two-parent recipient families.
• MFIP’s effects on marriage and on subsequent divorce among couples
who were cohabiting at study entry did not vary by race/ethnicity, age of
children, or prior employment or welfare history of the family. However,
MFIP cohabiting couples with three or more children were 52 percent less
likely than AFDC cohabiting couples ever to marry.
18
18
The increase in marriage among cohabiting recipients for families in which both parents had been em-
ployed prior to study entry (5.1 percentage points, or a 26 percent increase) is worth noting because of its magni-
tude; the same is true for cohabiting short-term welfare recipients (7.8 percentage points, or a 46 percent increase).
These effects may not have achieved statistical significance because of their small sample size, particularly for the
subgroup in which both parents had been previously employed. Effects on marriage among the subgroup in which
both parents had been employed prior to study entry overall (N=450) were statistically significant at 6.7 percent-
age points.
-18-
MFIP's Long-Term Effects on Marriage and Divorce
Figure 4
Cumulative Percentage Ever Married for Two-Parent Recipient Families Who Were Cohabiting
at Baseline, over a Seven-Year Follow-Up Period
-10
-5
0
5
10
15
20
25
Months after Random Assignment
Percentage
2 6 12 18 24 30 36 42 48 54 60 66 72 78 84
Impact
MFIP
AFDC
SOURCES: MDRC calculations using public divorce and marriage certificate records from the State of Minnesota.
N
OTES: The sample includes members randomly assigned from April 1, 1994 to March 31, 1996, excluding the small
percentage who were receiving or applying only for food stamps at random assignment. Approximately 96 percent of two-
parent recipient families were randomly assigned from April 1994 to March 1995.
Marriage outcome includes any match with the marriage certificate record over time. The analysis does not adjust for
divorce and excludes marriages to individuals other than the partner identified at study entry.
-19-
MFIP's Long-Term Effects on Marriage and Divorce
Figure 5
Figure 5
Marital Status in Each Month of Follow-Up for Two-Parent Recipient Families
Who Were Cohabiting at Baseline
-10
-5
0
5
10
15
20
25
Months after Random Assignment
Percentage Married
MFIP
AFDC
Impact
SOURCES: MDRC calculations using public divorce and marriage certificate records from the State of Minnesota.
N
OTES: The sample includes members randomly assigned from April 1, 1994 to March 31, 1996, excluding the small
percentage who were receiving or applying only for food stamps at random assignment. Approximately 96 percent of two-
parent recipient families were randomly assigned from April 1994 to March 1995.
Marriage outcome adjusts for documented divorces based on divorce records data. The analysis excludes marriages
to individuals other than the partner identified at study entry.
2 6 12 18 24 30 36 42 48 54 60 66 72 78 84
-20-
Cumulative Percentage Ever Divorced for Two-Parent Recipient Families Who Were Cohabiting
at Baseline, over a Seven-Year Follow-Up Period
MFIP's Long-Term Effects on Marriage and Divorce
Figure 6
-10
-5
0
5
10
15
20
Months After Random Assignment
Percentage
AFDC
Im
p
act
MFIP
2 6 12 18 24 30 36 42 48 54 60 66 72 78 84
SOURCES: MDRC calculations using public divorce and marriage certificate records from the State of Minnesota.
N
OTE: The sample includes members randomly assigned from April 1, 1994 to March 31, 1996, excluding the small
percentage who were receiving or applying only for food stamps at random assignment. Approximately 96 percent of two-
parent recipient families were randomly assigned from April 1994 to March 1995.
-21-
Table 5 presents MFIP’s effects on divorce and on ever being married among cohabit-
ing couples at study entry across various subgroups (those with adequate sample sizes). MFIP
cohabiting recipient couples with three or more children were 12.6 percentage points, or 52 per-
cent, less likely to marry, compared with AFDC cohabiting recipient couples with three or more
children. In contrast, MFIP cohabiting recipient couples with fewer than three children were
more likely to marry (by 7.9 percentage points, though not statistically significant). MFIP’s ef-
fects on marriage and on subsequent divorce among cohabiting recipient couples did not sig-
nificantly vary within any other subgroups. However, two noteworthy patterns did emerge.
First, MFIP’s effects on divorce among cohabiting recipient couples consistently occurred
across a variety of subgroups. Second, though effects on divorce were quite consistent across
subgroups, the pattern of effects on marriage varied across subgroups. For example, MFIP ap-
peared to reduce marriage and significantly reduced subsequent divorce among cohabiting re-
cipient couples who were long-term welfare recipients, but appeared to increase marriage with
no noted difference in subsequent divorce rate among cohabiting recipient couples who were
short-term welfare recipients. Future analyses of long-term effects on economic outcomes may
help explain some of these differing patterns in marriage and divorce across these subgroups.
Effects on two-parent applicant families
• MFIP had no average effect on divorce among two-parent applicant fami-
lies, but did somewhat increase divorce later in the follow-up period.
Table 1 shows that the rate of divorce over the seven-year follow-up period was similar
for MFIP and AFDC two-parent applicant families. However, MFIP two-parent applicant fami-
lies were slightly more likely to have divorced during the fourth through sixth year of follow-
up, with a statistically significant increase in divorce occurring in year five (see Table 2 and
Appendix Table 3).
• A trend showed increased marriage and increased subsequent divorce
among cohabiting applicant families. These effects, however, were not sta-
tistically significant during most years of the follow-up period, possibly
due to the small sample size of cohabiting applicant families.
Table 1 shows that for cohabiters, the pattern of results is different for applicants than
for recipient families. Compared with their control group counterparts, MFIP cohabiting appli-
cant families were more likely to be married by the end of the follow-up, though this pattern is
not statistically significant. MFIP cohabiting applicant families were also somewhat more likely
to ever divorce by the end of the follow-up. Although neither of these overall effects were
statistically significant over the seven-year follow-up period (as shown in Table 1), significant
P-Va
l
ue
f
o
r
P-Va
l
ue
f
o
r
Sample Percentage Subgroup Percentage Subgroup
Size MFIP AFDC Impac
t
a
Change Differences MFIP AFDC Impact
a
Change Differences
All recipient families 472 20.6 20.8 -0.3 -1.3 2.7 8.0 -5.3 ** -1.3
Number of children 0.01 ** 0.29
Fewer than 3 children 270 26.5 18.6 7.9 42.4 3.5 6.4 -3.0 -45.9
3 or more childre
n
195 11.7 24.3 -12.6 ** -51.9 2.0 10.1 -8.1 ** -80.1
Race/Ethnicit
y
White, non-Hispanic 308 25.9 23.6 2.3 9.8 4.0 9.9 -6.0 * -60.1
Age of youngest child
Less than 6 years old 428 21.1 22.0 -0.9 -4.0 2.6 8.4 -5.8 ** -69.4
Employment 1 yea
r
p
rior to study entr
y
0.63 0.82
One parent employed 196 25.9 30.5 -4.6 -15.2 3.8 9.4 -5.5 -59.1
Both parents employed 158 24.9 19.8 5.1 25.7 2.0 11.0 -8.9 ** -81.4
Welfare receipt prior to
study entr
y
0.23 0.17
Less than 2 years 153 24.8 17.1 7.8 45.7 6.3 7.2 -1.0 -13.2
More than 5 years 172 12.2 21.1 -8.9 -42.1 0.1 12.1 -12.0 *** -99.4
MFIP
'
s Lon
g
-Term E
ff
ects on Marr
i
a
g
e and D
i
vorce
Ever Married (%) Ever Divorced (%)
Table 5
Effects on Marria
g
e and Divorce for Two-Parent Recipient Families Who Were Cohabitin
g
at Baseline,
over a
S
even-Year Follow-Up Per
i
od
SOURCES: MDRC calculations using public divorce and marriage certificate records from the State of Minnesota.
N
OTE: The sample includes members randomly assigned from April 1, 1994 to March 31, 1996, excluding the small percentage who were receiving or applying only
for food stamps at random assignment. Approximately 96 percent of two-parent recipient families were randomly assigned from April 1994 to March 1995.
A two-tailed t-test was applied to regression-adjusted impact estimates. Statistical significance levels are indicated as ***= 1 percent; **= 5 percent; * =10
percent.
Rounding may cause slight discrepancies in sums and differences.
Information at baseline on some subgroup characteristics was missing for some sample members. Therefore, the average impact across subgroups does not
always replicate the impact for all recipients.
Subgroups with fewer than 150 families are not reported.
a
The difference is the impact of the financial incentives, mandatory services, reinforced incentive messages, and elimination of the 100-hour rule and work
history requirement.
-22-
-23-
increases in divorce did occur for cohabiting applicant families during the first two years of fol-
low-up (see Appendix Table 3).
• Although there were no effects on divorce among all married two-parent
applicant families, MFIP increased divorce among a few subgroups of
two-parent applicant families.
Appendix Table 4 presents findings on MFIP’s effects on divorce among subgroups of all
two-parent applicant families and subgroups of married applicant families.
19
MFIP increased di-
vorce by 11.6 percentage points among two-parent applicant families with three or more children
at study entry, an effect that appears to be clustered among those two-parent applicants with three
or more children who were married at study entry (see bottom panel of Appendix Table 4). This
effect on divorce is significantly different from MFIP’s effects on divorce among two-parent ap-
plicant families with fewer than three children. MFIP also increased divorce by 10 percentage
points among two-parent applicant families in which only one parent was employed prior to study
entry. However, this effect was not significantly different from MFIP’s effects on divorce among
two-parent applicant families in which both parents were employed prior to study entry.
Conclusions
In the original MFIP evaluation, effects of the program on marital stability were meas-
ured for 290 two-parent recipient families who were respondents to a 36-month follow-up sur-
vey. The new analysis presented here has expanded on these findings by providing long-term
follow-up for the full sample of 1,515 two-parent recipient families and 731 two-parent appli-
cant families who participated in the MFIP evaluation. Overall, the results indicate that the pilot
MFIP program that began in 1994 continued to have effects on rates of divorce for two-parent
families seven years after they entered the study. Reductions in divorce were concentrated
among families who were already receiving welfare when they entered the study rather than
new welfare applicants. Impacts were most pronounced for black parents who were already
married at the time they entered the study, and for parents who were cohabiting when they en-
tered the study.
Note that because this new analysis relies solely on public records of marriages and fi-
nalized divorces, it does not capture any effects that the program may have had on couples’ like-
lihood of separating or living apart without formally divorcing. Nevertheless, the findings rep-
resent some of the best evidence to date about the potential for welfare policies to affect marital
stability among two-parent families.
19
The sample of two-parent cohabiting couples was too small to pursue comparable subgroup analyses.
-24-
The results raise several important questions for future work. First, because most wel-
fare reform evaluations in the 1990s did not collect information on two-parent families, there is
little impact information available for two-parent families, making it important to replicate this
type of evaluation in some additional geographic areas to determine whether they are generaliz-
able. Second, to further understand the results presented here, future work will examine how
MFIP affected the timing of marriages or divorces, particularly for cohabiting parents. In addi-
tion, the suggestion of increased divorce among two-parent applicant families in certain years of
follow-up is worthy of further investigation.
Finally, the effects on divorce among MFIP’s two-parent recipient families suggest that
the program could have had important effects on the well-being of children in these families. A
finalized divorce is almost always preceded by marital conflict and a period of separation, both
of which are likely to extend over a long period of time. The effects on divorce presented here
very likely understate the duration and extent of marital strife imposed upon the children in
these families. With funding from various foundations, via administrative records (including
child welfare records), MDRC is continuing to conduct long-term analyses on the effects of the
pilot MFIP on economic, family, and child outcomes among single-parent and two-parent fami-
lies, with the expectation that additional findings will become available in 2004.
Appendix
-26-
Selected Characteristics of Two-Parent Sample Members,
b
y
Welfare Status at Random Assi
g
nment
Characteristic Recipients Applicants
Demographic characteristics
Geographic area (%)
Hennepin County (Minneapolis) 52.4 37.2
Anoka/Dakota counties 20.3 24.2
Rural counties 27.3 38.6
Gender of respondent (%)
Female 90.7 78.0
Male 9.3 22.0
Average age (years) 31.2 30.6
Race/Ethnicity (%)
White, non-Hispanic 59.5 79.7
Black, non-Hispanic 16.2 7.2
Hispanic 2.7 4.3
Native American/Alaskan Native 5.6 2.2
Asian/Pacific Islander 16.0 6.6
Family status
Marital status (%)
Never married 24.2 17.4
Married, living with spouse 68.7 78.8
Married, living apart 1.8 0.7
Separated 0.2 0.1
Divorced 5.2 3.0
Widowed 0.1 0.0
Age of youngest child in years (%)
Under 3, or client pregnant at the
time of random assignment 55.2 61.1
3-5 22.3 12.8
6-18 22.5 26.1
Number of children (%)
One 20.8 39.4
Two 31.1 28.1
Three or more 46.3 30.7
Labor force status
Worked full time for 6 months
or more for one employer (%) 52.4 73.5
Any earnings in past 12 months (%) 59.2 21.2
Currently employed (%) 15.1 30.6
Average hourly wage
a
($) 6.41 7.38
Average hours worked per week
b
(%)
1-19 36.7 36.4
20-29 24.3 15.5
30 or more 38.9 48.2
Never worked (%) 16.6 3.6
(continued)
A
ppend
i
x Table
1
-27-
Appendix Table 1 (continued)
Characteristic Recipients Applicants
Education status
Highest credential earned (%)
GED certificate
c
12.6 10.4
High school diploma 38.9 51.1
Technical/2-year college degree 9.2 12.7
4-year college degree or higher 2.1 7.0
None of the above 37.2 18.9
Highest grade completed in school (average) 11 12
Prior welfare receipt
Total prior AFDC receipt
d
(%)
None 3.7 74.4
Less than 4 months 4.5 2.6
4 months or more but less than 1 year 13.0 8.7
1 year or more but less than 2 years 13.4 4.3
2 years or more but less than 5 years 30.5 6.4
5 years or more but less than 10 years 23.0 2.4
10 years or more 12.1 1.2
Housin
g
status
Current housing status (%)
Public housing 7.6 2.1
Subsidized housing 17.8 3.4
Emergency or temporary housing 3.7 3.4
None of the above 70.8 91.1
Number of moves in the past 2 years (%)
None 34.8 34.2
1 or 2 45.6 50.1
3 or more 19.6 15.8
Current and recent education and trainin
g
activities
Currently enrolled in education or training
e
(%)
Any type 20.3 12.3
GED preparation 2.6 0.7
English as a Second Language 5.7 1.6
Adult Basic Education 1.2 0.6
Vocational education/skills training 4.5 2.2
Post-secondary education 3.4 4.0
Job search/job club 3.0 2.1
Work experience 0.5 0.4
High school 0.7 1.5
Enrolled in any type of education or training
during the previous 12 months (%) 28.4 16.0
Sample size (total = 2,256) 1,523 733
(continued)
-28-
Appendix Table 1 (continued)
SOURCE: Calculations using data from Background Information Forms.
N
OTES: The sample includes AFDC and MFIP group members who were randomly assigned
from April 1, 1994 to March 31, 1996, excluding the small percentage who were receiving or
applying only for food stamps at random assignment.
This table includes eight recipient families and two applicant families who were excluded
from the rest of the analysis because they did not provide information at baseline on their marital
status.
a
Calculated for those employed at the time of random assignment who reported an hourly
wage. Twenty percent of those employed were excluded because they did not report an hourly
wage.
b
Calculated for those employed at the time of random assignment.
c
The General Educational Development (GED) certificate is given to those who pass the
GED test and is intended to signify knowledge of basic high school subjects.
d
This refers to the total number of months an individual or her spouse has spent on AFDC at
one or more periods of time as an adult. It does not include AFDC receipt under a parent's name.
e
Totals may not equal all categories summed because some sample members may be in more
than one category.
-29-
Sample Percentage Percentage
Size MFIP AFDC Impact
a
Change MFIP AFDC Impact
a
Change
Full recipient sample 1,515
Year 1 0.5 0.5 0.0 0.7
Year 2 2.0 2.7 -0.8 -28.3
Year 3 3.3 4.8 -1.4 -30.2
Year 4 6.3 6.8 -0.4 -6.6
Year 5 7.5 9.1 -1.6 -17.1
Year 6 8.7 11.7 -3.0 ** -25.9
Year 7 9.1 12.9 -3.8 ** -29.5
Among those who were
married at baseline 1,043
Year 1 0.6 0.7 -0.1 -16.6
Year 2 2.5 3.6 -1.2 -32.2
Year 3 4.3 6.0 -1.7 -28.1
Year 4 8.6 8.8 -0.3 -3.0
Year 5 10.2 12.0 -1.8 -15.1
Year 6 11.7 15.0 -3.3 -21.7
Year 7 12.1 16.8 -4.6 ** -27.7
Among those who were
cohabiting at baseline 472
Year 1 0.2 0.0 0.2 N/A 6.4 7.2 -0.8 -11.0
Year 2 0.9 0.8 0.1 10.0 11.0 11.0 0.0 0.0
Year 3 0.8 2.5 -1.7 -67.4 13.8 15.8 -2.0 -12.9
Year 4 0.8 3.4 -2.6 * -76.2 16.1 17.1 -1.0 -6.0
Year 5 1.1 4.0 -2.9 * -72.4 17.5 18.2 -0.7 -3.9
Year 6 1.7 5.5 -3.8 ** -68.9 18.6 17.3 1.4 7.8
Year 7 2.4 4.6 -2.2 -48.0 20.4 15.9 4.5 28.2
{ Not applicable }
{
Not applicable
}
Appendix Table 2
Effects on Marriage and Divorce for Two-Parent Recipient Families
over a Seven-Year Follow-Up Period, Year by Year
Divorced at End of Year (%) Married at End of Year (%)
SOURCES: Calculations using public divorce and marriage certificate records from the State of Minnesota.
N
OTES: The sample includes members randomly assigned from April 1, 1994 to March 31, 1996, excluding the small
p
ercentage who were receiving or applying only for food stamps at random assignment. Approximately 96 percent of
two-parent recipient families were randomly assigned from April 1994 to March 1995.
A two-tailed t-test was applied to regression-adjusted impact estimates. Statistical significance levels are
indicated as ***= 1 percent; **= 5 percent; * =10 percent.
Rounding may cause slight discrepancies in sums and differences.
Because of the long intake period, the full sample size was not available for follow-up Year 7; the sample size at
the end of Year 7 was 1,235.
a
The difference is the impact of the financial incentives, mandatory services, reinforced incentive messages, and
elimination of the 100-hour rule and work history requirement.
-30-
Sample Percentage Percentage
Size MFIP AFDC Impact
a
Change MFIP AFDC Impact
a
Change
Full applicant sample 731
Year 1 0.6 1.7 -1.1 -63.3
Year 2 3.1 2.9 0.2 7.2
Year 3 5.7 6.8 -1.1 -16.3
Year 4 12.3 8.9 3.5 39.1
Year 5 16.2 11.9 4.3 * 35.8
Year 6 19.0 15.8 3.3 20.
6
Among those who were
married at baseline 577
Year 1 0.5 2.1 -1.6 -74.5
Year 2 4.1 3.4 0.7 19.6
Year 3 7.2 8.0 -0.8 -9.7
Year 4 15.0 10.5 4.5 42.7
Year 5 19.8 14.2 5.6 * 39.1
Year 6 23.2 19.2 3.9 20.5
Among those who were
cohabiting at baseline 154
Year 1 1.9 0.0 1.9 * 0.0 7.9 4.3 3.6 82.2
Year 2 1.9 0.0 1.9 * 0.0 13.8 10.2 3.6 35.2
Year 3 0.5 3.0 -2.5 -82.2 16.0 10.7 5.3 49.2
Year 4 3.8 2.7 1.0 37.0 19.5 12.5 7.0 55.4
Year 5 4.4 4.0 0.4 10.4 20.1 13.8 6.3 46.0
Year 6 4.9 4.3 0.6 15.1 21.5 15.0 6.6 44.0
{
Not applicable
}
{
Not applicable
}
Appendix Table 3
MFIP's Effects on Marriage and Divorce for Two-Parent Applicant Families
over a Six-Year Follow-Up Period, Year by Year
Divorced at End of Year (%) Married at End of Year (%)
SOURCES: Calculations using public divorce and marriage certificate records from the State of Minnesota.
N
OTES: The sample includes members randomly assigned from April 1, 1994 to March 31, 1996, excluding the
small percentage who were receiving or applying only for food stamps at random assignment.
A two-tailed t-test was applied to regression-adjusted impact estimates. Statistical significance levels are
indicated as ***= 1 percent; **= 5 percent; * =10 percent.
Rounding may cause slight discrepancies in sums and differences.
Because of the long intake period, the full sample size was not available for follow-up Year 6; the sample
size at the end of Year 6 was 696.
a
The difference is the impact of the financial incentives, mandatory services, reinforced incentive messages,
and elimination of the 100-hour rule and work history requirement.
-31-
P-Value for
Sample Percentage Subgroup
Size MFIP AFDC Impact
a
Change Differences
Overall
All applicant families 731 21.0 18.8 2.2 11.7
Number of children 0.08 *
Fewer than 3 children 494 17.3 18.6 -1.3 -7.1
3 or more children 224 31.3 19.7 11.6 * 58.8
Race/Ethnicit
y
0.16
White, non-Hispanic 575 21.9 21.0 0.8 3.9
Age of youngest child 0.21
Less than 6 years old 530 23.6 19.7 3.8 19.3
6 years old or older 188 13.2 17.8 -4.7 -26.3
Employment 1 year prior to study entr
y
0.12
One parent employed 226 24.4 14.3 10.1 * 70.6
Both parents employed 373 20.2 24.8 -4.6 -18.6
Welfare receipt prior to study entr
y
Less than 2 years 650 21.7 19.1 2.5 13.2
Married at Baseline
All applicant families 577 25.0 23.5 1.5 6.3
Number of children 0.09 *
Fewer than 3 children 362 21.3 24.2 -2.9 -11.9
3 or more children 208 33.0 21.8 11.1 50.9
Race/Ethnicit
y
White, non-Hispanic 454 25.2 26.9 -1.7 -6.4
Age of youngest child 0.40
Less than 6 years old 388 28.7 27.0 1.7 6.4
6 years old or older 182 13.7 18.5 -4.8 -26.0
Employment 1 year prior to study entr
y
0.10 *
One parent employed 191 27.2 16.0 11.2 70.4
Both parents employed 277 25.9 33.6 -7.7 -22.9
Welfare receipt prior to study entr
y
Less than 2 years 512 25.3 23.6 1.8 7.5
MFIP's Long-Term Effects on Marriage and Divorce
Ever Divorced (%)
Effects on Divorce for Two-Parent Applicant Families
over a Six-Year Follow-Up Period
Appendix Table 4
SOURCES: MDRC calculations using public divorce and marriage certificate records from the State of Minnesota.
N
OTES: The sample includes members randomly assigned from April 1, 1994 to March 31, 1996, excluding the
small percentage who were receiving or applying only for food stamps at random assignment.
A two-tailed t-test was applied to regression-adjusted impact estimates. Statistical significance levels are
indicated as ***= 1 percent; **= 5 percent; * =10 percent.
Rounding may cause slight discrepancies in sums and differences.
a
The difference is the impact of the financial incentives, mandatory services, reinforced incentive messages,
and elimination of the 100-hour rule and work history requirement.
-33-
References
Miller, Cynthia, Virginia Knox, Lisa A. Gennetian, Martey Dodoo, Jo Anna Hunter, and Cindy Red-
cross. 2000. Reforming Welfare and Rewarding Work: Final Report on the Minnesota Family
Investment Program, Volume 1: Effects on Adults. New York: MDRC.
Knox, Virginia, Cynthia Miller, and Lisa A. Gennetian. 2000. Reforming Welfare and Rewarding
Work: A Summary of the Final Report on the Minnesota Family Investment Program. New
York: MDRC.
Harknett, Kristen, and Lisa A. Gennetian. In press. “How an Earnings Supplement Can Affect
Union Formation Among Low-Income Single Mothers.” Demography.
