The direct financial costs of having a family member incarcerated

Abstract

Using original data from the Family Incarceration Costs Survey, we present national estimates of the direct financial costs of family member incarceration. We find that most Americans with an incarcerated family member provide them direct financial support. The median monthly direct expense among those who contribute is $172, which represents 6% of their household income. On average, Blacks and Hispanics incur higher direct expenses than whites despite their lower household incomes. Men and women contribute similar amounts, but these direct expenses reflect a larger share of women’s household income. Poor families’ direct expenses are comparable to those of affluent families and are similar to their spending on health care, utilities, and car-related costs. Together, these results suggest that familial incarceration is a prominent line item that strains marginalized families’ already-tight household budgets and is a substantial yet underappreciated mechanism through which mass incarceration has reshaped the texture of American poverty in recent decades.

Many US adults with an incarcerated family member spend hundreds of dollars each month providing them direct financial support.

INTRODUCTION

Despite declines since 2008, the American incarceration rate remains comparatively extreme, historically unprecedented, and concentrated among men with low levels of education or from historically marginalized groups (1–4). The costs of this extent of incarceration have mostly received attention in three areas: (i) the financial costs to taxpayers (5); (ii) the ramifications for incarcerated and formerly incarcerated people in the form of fines and fees, suffering during incarceration, and life chances after release (6–14); and (iii) the social costs for women and children, ranging from housing instability to lower educational attainment to worse health (15–21).

Although some of the earliest work on the consequences of incarceration focused on the financial costs to individuals with incarcerated family members (22), little recent work has quantified these costs. This is unfortunate for three reasons. First, earlier studies produced sizable estimates of these costs (22, 23). The initial study in this area, for instance, indicated that in a sample of Black women visiting loved ones in a facility in California, the poorest women spent 26% of their income visiting, calling, and sending packages to their incarcerated loved one (16).

Second, few basic necessities are provided by prisons and jails, and departments of corrections often mark up items that incarcerated persons can purchase. Incarcerated individuals almost universally report that food in prisons and jails is unappetizing and leaves them hungry (9, 24). Ramen noodles, a cheap and effective way to satiate this hunger, costs about 35 cents per pack on the outside; in many prisons and jails, ramen costs two to three times that amount (25). This problem is compounded by incarcerated individuals having little opportunity to earn money: The most recent data suggest that the maximum pay rate for an incarcerated person in a Florida prison is 55 cents per hour (26), while Florida prisons charge $1.09 per ramen packet—meaning that it would take 2 hours of work at the highest earning rate to purchase one ramen packet (25). On the outside, even at the federal minimum wage of $7.25, it would take a mere 3 min to earn the necessary 35 cents to buy a ramen packet. This combination of meager pay, insufficient provision of necessities, and high markups leaves incarcerated people sorely reliant on family members to meet their basic needs via the types of contributions we focus on here.

Third, the families of the incarcerated are often economically disadvantaged such that the costs of family member incarceration may exacerbate existing hardship. Ethnographic research paints a portrait of prior deep deprivation that is intensified during incarceration (24, 27–29). Quantitative research shows that families were often struggling before incarceration and experienced an uptick in hardships, such as having utilities turned off during incarceration (30). Therefore, overlooking these costs risks overlooking a key vehicle through which mass incarceration has exacerbated social stratification in recent years.

This article introduces an original dataset, the Family Incarceration Costs Survey (FICS), which uses NORC at University of Chicago’s AmeriSpeak panel to provide nationally representative estimates of both direct and indirect costs borne by American adults who have an incarcerated family member. Moreover, the FICS allows us to disaggregate such estimates across prisons versus jails, race and ethnicity, gender, and family member relationship type. By providing such estimates, we offer needed insight into who contributes, how much they contribute, and how their contributions compare to their household income and common expenditures like groceries and health care.

The direct expenses analyzed in this article include three distinct types of provision: paying for collect calls, putting money on their incarcerated family member’s commissary account (often referred to colloquially as putting money “on their books”), and designated financial support for goods and services that may occur in place of, or in addition to, commissary payments. Of course, these are not the only potential costs that could be associated with family member incarceration—families may visit their incarcerated loved one, move to new dwellings, support them in the immediate reentry period, or help pay their legal fines and fees, to name but a few possibilities. Therefore, we also lean on related work in this area—including estimates from a recent policy report that used FICS data to generate an estimate of the population-level total cost of familial incarceration (31)—to conduct a back-of-the-envelope aggregated calculation incorporating these more indirect or tangential costs at the conclusion of the article.

We see our focus here on direct recurring expenses as distinctly meaningful for three reasons. First, these are expenses that are typically borne at regular intervals because they are essential to the basic needs and well-being of the incarcerated individual. Second, these costs are transactional in the sense that they represent a direct transfer between an individual and their incarcerated family member—and one that exists within the context of a uniquely one-sided dynamic, given the incarcerated person’s inability to generate meaningful income or provide in-kind forms of support for those on the outside while incarcerated. Last, because of the recurring and essential nature of these items, they may often become line items in one’s budget and could subsequently squeeze out necessary spending in other domains.

RESULTS

Table 1 presents estimates of (i) the proportion of dyads in the FICS with any reported direct recurring expenses and (ii) median contributions and median proportions of household income contributed both for the full sample and among those incurring any cost. Note that all analyses are conducted at the dyad level, meaning that they are between a given respondent and each incarcerated family member. Thus, to the extent that some adults have multiple family members incarcerated simultaneously, their aggregate expenditure will be greater than the dyadic-level amounts presented here (also see Materials and Methods).

Table 1. Overview of sample characteristics and expenditures.

The two leftmost columns show the number of dyads N for each subgroup, i.e., the analytic sample posttruncation, along with the proportion of dyads within each subgroup that incurred any direct expense. The remaining four columns show the median expenditure amount and median proportion of household income for each subgroup and provide separate estimates for the entire sample versus those who incurred any direct expense. Note that all expenditure amounts are estimated at the dyad level, i.e., between a given respondent and one incarcerated family member.

			Median direct expenses (monthly $)		Median proportion of monthly household income spent on direct expenses
	No. of dyads	Proportion incurred any direct cost	Full sample	Among those who incurred expense	Full sample	Among those who incurred expense
All	1880	0.64	36	172	0.01	0.06
Facility
Jail	711	0.58	16	166	0.01	0.06
Prison	1062	0.71	63	175	0.02	0.06
Gender
Male	764	0.61	19	150	0.01	0.04
Female	1116	0.66	45	185	0.02	0.08
Race/ethnicity
White	876	0.59	11	120	0.00	0.04
Black	500	0.71	83	200	0.04	0.09
Hispanic	358	0.65	61	230	0.03	0.09
Native American	55	0.58	1	75	0.00	0.05
Relationship
Coparent/spouse	498	0.67	102	276	0.04	0.12
Mother	45	0.73	82	286	0.04	0.08
Father	189	0.60	17	176	0.01	0.08
Child	256	0.73	84	180	0.03	0.06
Sibling	780	0.60	20	125	0.01	0.04

The results indicate that most dyads incurred at least one of these three direct expenses (collect calls, commissary, and designated giving), with nearly two-thirds of dyads reporting at least one form of contribution (64%). Differences in propensity to incur any direct expense across facility type, gender, race and ethnicity, and family member relationship type were small, with the range across these groupings spanning from about 58% (for Native Americans) to 73% (when the incarcerated family member is either one’s child or mother). Thus, most dyadic pairs make some contribution to the incarcerated family member via the channels we consider here.

The median contribution among those who incurred any direct expenses was $172 per month. Of those who incurred any expense, differences in the median expenditure were substantial for Blacks and Hispanics ($200 and $230 per month, respectively) versus whites ($120 per month). Relative to other relationship types, median expenses among those who contributed were especially high for those whose coparent/spouse or mother was incarcerated, at $276 and $286 per month, respectively. Differences in median expenses were smaller for men relative to women ($150 to $185 per month) and between prisons and jails ($175 to $166 per month).

Among dyads in the FICS who incurred any direct expenses, the median percentage of the respondent’s household income contributed was 6%. Median percentages of household income expended among those who contributed anything were high for Blacks (9%), Hispanics (9%), and those who had a spouse or coparent incarcerated (12%). For whites, the median share of household income among those who contributed was lower at 4%; the median male contributed less of his household income than the median female (4 and 8%).

Thus, Table 1 provides evidence that the median contribution for family members who do contribute is substantial ($172 per month) and that the primary source of stratification in the amount of direct expenses is across racial and ethnic lines, with Blacks and Hispanics incurring higher expenses than whites—both in absolute terms and as a proportion of household income. This is especially noteworthy because Blacks and Hispanics experience higher rates of family member incarceration than whites (32–34).

Figure 1 presents box and whisker plots of the amount spent per month by contributors on these direct expenses and includes estimates of mean contributions alongside bootstrapped confidence intervals, both for the full sample and for those who contributed anything. Table S1 provides exact numbers for each statistic in the plot.

Fig. 1. Distribution of monthly direct expenses by subgroup.

The top and bottom of the box represent the 25th (Q1) and 75th (Q3) percentiles, respectively, while whiskers extend toward zero as calculated by Q1 − 1.5 * IQR and toward the maximum value as calculated by Q3 + 1.5 * IQR. Outliers (shown as dots) are values that extend beyond the whiskers. Box and whisker calculations are among contributors only; however, within each box are two horizontal lines representing the mean values: one for the mean of the full sample (in faded color) and one for those who incurred any cost (in dark color). Bootstrapped confidence intervals for each mean are also shown by the vertical bars overlapping their corresponding mean value. For visual clarity, the y axis was restricted to a maximum of 3000, which visually omits one observation of $3409.

For the full sample, which includes those who did not contribute, the mean monthly expenditure was $208; when considering only those who had contributed, the mean monthly expenditure was $331. If those levels of monthly contributions held over an entire year, the mean levels of spending would be $2496 for the entire US adult population and $3972 when limited to those who contributed [note that the latter number is slightly different from the equivalent by Elderbroom et al. (31) because of rounding]. Recall that these estimates are only at the dyad (i.e., relationship) level and therefore may underestimate the full extent of expenditures among individuals who support multiple incarcerated family members simultaneously. Differences in contributions for individuals who had a family member incarcerated in a jail or in a prison were small and not significant—the same was true for males and females.

Mean contributions were larger for Blacks ($280 including those who did not contribute and $413 excluding those who did not) and Hispanics ($225 and $365) than for whites ($152 and $252), although only the Black-white difference was statistically meaningful. Distributionally, we can also see visually that the upper range (i.e., the top of the boxes and the upper whisker) of expense amounts is greater for Blacks and Hispanics.

The figure also suggests that incarcerated spouses or coparents receive more from their partner than any other family member type ($306 and $486), except incarcerated mothers (which is a small enough group that there is much uncertainty in our estimates for them). Visually, we observe that the contributor mean for coparent/spouse is above the 75th percentile (i.e., the upper limit of the box) for fathers, children, and siblings.

Stretched across a year, the direct expenses incurred by Blacks and individuals (of any race) with a spouse or coparent incarcerated would translate to yearly rates of $3360 (for the full sample) and $4956 (among those who contributed) for Blacks and $3672 (for the full sample) and $5832 (among those who contributed) for dyads in which the incarcerated family member was one’s spouse or coparent.

Figure 2 presents analogous estimates to Fig. 1 but for the proportion of self-reported household income during the time of their family member’s incarceration. For the full sample, which includes individuals who did not contribute to these expenses, the mean percentage of household income spent per month was 12%; for those who contributed, the corresponding number was 18%. This suggests that among dyads where any expense was incurred, 18% of household income was dedicated to covering these recurring costs. Put differently, this result highlights that individuals who contributed in any of these domains on behalf of their incarcerated family member spent nearly one-fifth of their household income doing so.

Fig. 2. Distribution of monthly expenses incurred as a proportion of household income by subgroup.

The top and bottom of the box plots represent the 25th (Q1) and 75th (Q3) percentiles, respectively, while whiskers extend toward zero as defined by Q1 − 1.5 * IQR and toward the maximum value as defined by Q3 + 1.5 * IQR. Outliers (shown as dots) are values that extend beyond the whiskers. Box and whisker calculations are among contributors only; however, within each box are two horizontal lines representing the mean values: one for the mean of the full sample (in faded color) and one for those who incurred any cost (in dark color). Bootstrapped confidence intervals for each mean are also shown by the vertical bars overlapping their corresponding mean value. We top-code proportions to 1 in cases where a respondent provided an expense amount greater than their reported income.

Differences were again quite pronounced across race and ethnicity. Blacks (18% of household income in the full sample and 25% excluding those who did not contribute) contributed a significantly larger share of their household income than did whites (8 and 13%). Individuals with a spouse or coparent incarcerated contributed the most of any group at 18 and 27% of their household income. Women with a family member incarcerated (14 and 21%) also contributed more of their income than men (10 and 15%), even though their amount of monthly giving was similar. Differences across facility type were again not significant.

Figure 3 provides information on how this spending compares to spending in other domains using data from the 2022 Consumer Expenditure Survey (CES; see Materials and Methods for details), broken down by household income level. Note that both the CES and family incarceration cost data in Fig. 3 represent mean values and that the family incarceration costs are still at the dyad level. For adults from the poorest households (defined as those with household incomes under $15,000), Fig. 3 shows that the average monthly spending on direct incarceration expenses was just over $200, roughly what individuals in these households spent on health care, utilities, and car insurance and gas.

Fig. 3. Average monthly direct expenses for family member incarceration compared to select household expenses, stratified by household income level.

We combine our estimates with data on common types of personal expenses from the 2022 CES, run by the US Bureau of Labor Statistics, to provide a comparison of such costs among each comparable income group between the CES and the FICS. All amounts shown are the mean values, with the family incarceration cost estimate a dyad-level weighted mean among all respondents within a given income category (analogous to the faded mean line in Fig. 1). A bootstrapped confidence interval for the family incarceration cost estimate is also displayed. The dashed line represents the weighted mean for the family incarceration cost estimate over everyone in the full sample. Note that the FICS measures of income are $50,000 to $74,999 and $75,000 to $99,999 compared to the ranges defined in the CES and used in the figure above: $50,000 to $69,000 and $70,000 to $99,999.

As household income increases, two trends merit attention. First, there is no statistically or substantively significant increase in incarceration-related direct expenses for higher-income individuals. In fact, one of the higher-income groups ($70,000 to $99,999) is the only group whose incarceration expenses are statistically lower than the overall mean. Second, spending in other domains increases steadily across other categories as household income increases, yet family incarceration expenses appear far more inelastic and remain virtually unchanged. While acknowledging that families in the CES may be qualitatively different (even within-income group) from FICS respondents in ways that alter their spending, it remains descriptively clear that family incarceration expenses constitute a much larger share of poor families’ budgets. This may well impede their ability to spend on these other goods and services—a causal inquiry we cannot test but hope is taken up by future research.

Last, in Fig. 4, we further quantify the uncertainty around measurement error and our own decisions for handling this source of variation. Figure 4 visualizes (separately for the full sample versus contributors only) the mean value for each expenditure category based on one-percentage increments of truncation ranging from the 100th percentile (no truncation) to the 90th percentile. Note that we do this with means given their inherent sensitivity toward outliers (see Materials and Methods for a more robust discussion). Table S2 also provides uncertainty estimates when considering potential over- or underreporting tendencies of respondents.

Fig. 4. Mean expenditure at different truncation levels.

The x axis represents different truncation levels (the 90th percentile through the 100th percentile, indicating no truncation). See Materials and Methods for more details on decisions and calculations for truncation.

Visually in Fig. 4, we can observe that removing only the very largest outliers (moving from no truncation to truncating at the 99th percentile) results in the biggest decrease, followed by a more moderate decrease. This reinforces that even at a very heavy-handed truncation level of the 90th percentile, the total direct expenses would still be quite notable in terms of magnitude: more than $100 when averaged across the full sample of adults with an incarcerated family member and more than $200 among those who incur any direct expense.

DISCUSSION

Advocates, policymakers, and the media express increasing concern about the lack of basic necessities provided in prisons and jails and the often-exorbitant markups on such goods, as well as the high costs of keeping in touch with incarcerated loved ones (35, 36). Most recently, in May 2024, US Senators Cory Booker and Elizabeth Warren introduced a new act called the Families Over Fees Act with the goal of reducing surcharges and prices in facilities. Yet, research on how these costs cascade down to individuals who support their incarcerated family members is lacking. Here, we provide nationally representative estimates of the direct recurring expenses that American adults incur while they have a loved one incarcerated. In doing so, we illuminate the extent to which incarceration-related expenses carve into household budgets and examine how these burdens vary by sociodemographic groups.

Our results provide support for three conclusions. First, rates and amounts of contribution to incarcerated family members are high on average. Most (64%) individual-incarcerated family member dyadic pairs reported incurring at least one direct expense. Among those who did, the median contribution was $172 per month, representing 6% of their household income. This suggests that the direct expenses of having a family member incarcerated are substantively large.

Second, although rates of incurring any direct expense are similar across race and ethnicity, gender, and family member relationship type, the amount of such expenses differs substantially. Our data indicate that Blacks and Hispanics incur markedly higher expenses than whites—both in terms of dollars and in terms of the share of household income. Women and individuals with an incarcerated spouse or coparent (which are disproportionately women) also incur higher expenses than men and individuals with a nonspousal family member incarcerated, although this only becomes statistically relevant for women with respect to their share of household income.

Such disparities are noteworthy because they suggest that Blacks, Hispanics, and women—who already experience family member incarceration more frequently than other individuals—also experience the financial burden of family member incarceration per incarcerated family member more sharply. To underscore this point, FICS prevalence estimates show that in recent years, Blacks and Hispanics experience family member incarceration stints of 3+ months at 1.61 and 1.23 the rate of whites (see table S3). Such disparities in prevalence alongside the disparities at the dyad level that we document throughout this article suggest a compounding effect that carries profound and previously unidentified implications for our understanding of the proliferation and perpetuation of social inequality (33).

Yet, when looking only at the amount spent, we especially understate the cost for women and for Blacks because their household incomes on average are lower—a fact also highlighted in recent work on the implications of child incarceration for mothers’ wealth and assets (37). The resulting burden on women suggests that familial incarceration could be conceptualized as another “insult to health” described by Geronimus’s “weathering hypothesis,” which originally posited that the comparatively worse health of Black women was a physical byproduct of an accumulation of hardships experienced through early adulthood (38–40). The FICS results provide evidence of a mechanism that could lead to increased weathering of African-American women—and, likely, poor white women—given that these incarceration-related expenses compound on existing poverty and economic hardship. In turn, this may exacerbate already-marginalized women’s stress and negative coping behaviors while further limiting their access to health care.

Third, the direct expenses of having a family member incarcerated represent an important yet previously unmeasured budget line item for American families. Incarceration expenses can represent a sizable portion of the family budget, yet we find that incarceration expenses do not change proportionate to income—meaning that the burdens of incarceration fall especially on those already experiencing financial hardship. In addition, our estimates indicate that the poorest American households spend roughly the same amount per month per family member incarcerated as they do on other essentials like health care, utilities, and car insurance and gas. Meanwhile, middle-class and affluent Americans contribute no more than low-income households to incarceration-related expenses but spend far more on household goods and services. These findings are illustrative because they suggest that the poorest families may be forced to do without daily necessities partially because they have a family member incarcerated—a hardship that they also experience at a far greater rate than middle- and upper-income families.

Despite the importance of these findings, there are a number of limitations inherent in our data: Expense data are self-reported and sometimes from years in the past, which leaves them susceptible to misreporting and other biases (but see table S2); some outliers make it difficult to assess the reliability of means (but see Fig. 4); there is heterogeneity across facilities and states that we were unable to consider in our survey; costs may fluctuate over time and be subject to inflation; and all expenses are equalized to monthly despite the fact that some respondents indicated incurring expenses at frequencies more or less often than monthly. Materials and Methods contains thorough details of our strategies for addressing each of these limitations.

There are also a number of challenges and decisions—both conceptual and analytic—worth further contemplation. One is considering the counterfactual: Would there still be some financial transfer or support occurring between adult family members, especially in the case of a person who is otherwise marginalized but not removed from society through incarceration? Although we are unable to test this possibility directly with the data at hand and are not attempting to draw any causal conclusions from our analyses, existing research on the family lives of individuals who will go on to experience incarceration highlights that many family members were making substantial contributions to family life—in the form of direct financial contributions, shared expenses, and in-kind support such as providing childcare—before experiencing incarceration (27, 41–43).

Moreover, while incarcerated people may very well have had some prior earnings of their own—a construct that is not measured in the FICS but that recent research estimates to be about $5000 per year (44)—that could be used to pay for their own goods and services in jail or prison, the unique nature of the relationship between an incarcerated person and their nonincarcerated family member is necessarily one-directional during the incarceration period, in the sense that the incarcerated person is mostly prevented from earning any income and thus is unable to, for example, repay loans while incarcerated. As such, even if the net flow of resources is no different before and during the incarceration period, the nature of the relationship changes during incarceration, as the incarcerated individual has no way to repay the direct financial contributions their family members make, leading to these financial contributions representing not help that could be paid back in short order but something more properly conceived of as a gift that cannot be returned—as a debt that the family member will continue to owe (28, 45). It remains distinctly possible, however, that in some cases, there would still be financial costs incurred even in the absence of custodial sanctioning. It is also possible that these costs predated the incarceration period.

Another consideration is that our focus here is on direct recurring expenses or costs that represent a nonincarcerated person–to–incarcerated person transfer, which often occurs at relatively frequent and regular intervals. There are, however, many potential ways to conceptualize “costs” and many potentially less direct consequences that still bear financial ramifications for family members. In Table 2, we highlight a number of these indirect costs from a recent policy report using the FICS data (31).

Table 2. Direct and indirect costs from FICS data.

The estimates for visitation, moving, and reentry are reported by Elderbroom et al. (31).

Category	Mean cost among those who incurred	Source
Direct costs	$331 per month	Table S1
Visitation	$1703 per year	FWD p. 26
Moving	$2360 per move	FWD p. 20
Reentry	$905 per incarceration	FWD p. 20

To translate into a tangible (although admittedly imperfect) estimation, if we consider a hypothetical example in which one’s family member was incarcerated for 1 year, the family member was forced to move one time, and the family member did contribute directly, this would result in an aggregated financial cost from FICS data of $8940 (derived by annualizing the $331 mean monthly and combining with the remaining Table 2 estimates).

In addition, even this estimate neglects to consider the role that family members often play in paying fines and fees, which are not captured in the FICS given that they are not exclusively tied to incarceration. Prior research indicates median legal financial obligations ranging from $2700 among a sample of system-involved adults in Alabama (46) to $7234 among a sample of Washington state individuals convicted of a felony (7). Thus, if we extend the prior calculation to be specific to a felony conviction and assume that the family member paid 15% of their family member’s $7234 in fines/fees, this would push the estimate to more than $10,000 for just a single year-long stint of incarceration. [Note that to our knowledge, there is no prevailing national estimate of the average proportion of incarceration-specific fines and fees incurred by one’s family member at the dyad level, and thus for the purpose of this exercise, we chose a reasonably central figure in the most common range of estimated total contribution (0 to 24%; see table S6).]

Moreover, others have attempted to aggregate up these costs (and others) while accounting for the large number of people who experience family member incarceration, coming to a total yearly loss of $348 billion to affected families (31). Although there are many sources of uncertainty in estimates like these and many potential costs that are excluded or imprecisely measured, these figures highlight how high the family costs of incarceration can be above and beyond the direct recurring monthly expenses that we have focused on, and we believe that the field would do well to consider these more thoroughly in future research on the “ledger” of mass incarceration (47).

Limitations aside, the FICS data and findings presented here provide insight into the financial weight of mass incarceration for families and the unequal way in which it is experienced. Future research should seek to test (i) what factors drive family members to contribute to the financial costs associated with incarceration, (ii) how facility- and state-level variation in policies and procedures (such as upper limits on spending imposed by correctional facilities) influences the patterning of direct expenses, (iii) why the observed racial and ethnic disparities in financial costs exist, and (iv) whether these disparate burdens represent a core mechanism in the weathering of Black women in the era of mass incarceration.

The final two points are especially urgent, as there are plausible reasons to think that the differences we find here could represent differential familial ties and relationship dynamics, higher levels of unmet need for incarcerated Black and Hispanic individuals, a greater understanding of the needs of incarcerated individuals among Black and Hispanic families, or something else entirely. In addition, given that the prevalence of family member incarceration, propensity to incur direct expenses, and the amount of average direct expenses are all the highest among Black and Hispanic individuals, we contend that understanding the forces underlying these patterns—and their resultant implications for health, well-being, and financial hardship—is especially central for policymakers and scholars of social inequality.

MATERIALS AND METHODS

The lack of data suitable for estimating the direct recurring expenses associated with having a family member incarcerated necessitated collecting our own data. Therefore, we created the FICS.

The FICS was funded by FWD.us and administered by NORC at the University of Chicago using their AmeriSpeak Panel, which is a probability-based panel designed to be representative of the US household population. Randomly selected US households are sampled using area probability and address-based sampling, with a known, nonzero probability of selection from the NORC National Sample Frame. These sampled households are then contacted by US mail, telephone, and field interviewers (face to face). The panel provides sample coverage of ~97% of the US household population. While most AmeriSpeak households participate in surveys by web, noninternet households can participate in AmeriSpeak surveys by telephone. Households without conventional internet access but having web access via smartphones are allowed to participate in AmeriSpeak surveys by web. See the Supplementary Materials for additional description.

Sample inclusion

The target population is individuals aged 18+ with at least one family member who has been incarcerated for at least 3 months since 2016. Of the 31,450 panel members invited to take the survey, 1604 individuals completed the survey after qualifying on the basis of the screener questions. Respondents were screened into the survey on the basis of two questions: first, whether they had ever had an immediate family member incarcerated, and second, if so, whether that family member had been incarcerated for at least 3 months since 2016. While secondary in terms of the project’s goal, this also enables us to provide prevalence estimates for recent and substantial family member incarceration, the results for which we show in table S3. Direct comparisons to other prevalence estimates are difficult given the uniqueness of our target population, and we intend for our results to be a potential benchmark for future research in this area. Nonetheless, we still undertake a series of benchmarking exercises in the Supplementary Materials (tables S3 to S5), which reinforce our sampling and survey process.

Consistent with previous research (33), we defined an immediate family member as including one’s spouse, anyone they have had a child with, mother, father, sister, brother, child, granddaughter, or grandson—including step, foster, and adoptive family members. The second and more restrictive inclusion criterion was chosen for two reasons: (i) Many adults end up in jail for short periods of time, often less than 24 hours, which would not incur any of the direct recurring expenses that we quantify in this manuscript, and (ii) we wanted the incarceration periods to be recent, both to account for rising costs/inflation and to limit recall bias. The year 2016 was settled on because of it being an election year, which would be salient from a memory perspective.

Eligible respondents were able to indicate whether they had 1, 2, or 3+ immediate family members who met our inclusion criteria. For those with one or two immediate family members that met the criteria, we asked respondents to answer the entire series of questions about each family member. For those with 3+ immediate family members, we asked them to answer for two family members, which were selected with the following prompt: “please think about your incarcerated family member whose first name comes first in the alphabet.” This left us with data on 2118 dyadic family-member relationships from the overall qualified sample. These dyads form the basis of our analyses.

We conduct analyses at the dyad level given that we do not have sufficient information to calculate a person-level or household-level estimate (which would require complete cost estimates for every incarcerated family member of a given respondent and the total number of family members that a respondent has, regardless of their incarceration status), and even if we did have this information, we believe that the relationship-level estimates are most meaningful for two reasons: They (i) provide a statistic that is not predicated on the size of one’s family or the number of incarcerated family members that one has, which may or may not include multiple incarceration stints that overlap with one another, and (ii) enable disaggregation of expenses along two dimensions—facility type or incarceration duration (jail versus prison) and familial relationship (mother, brother, spouse, etc.) that we believe are vital for understanding the nuanced dynamics of this issue.

Upon reviewing the data, additional cleaning was clearly needed. Data for five respondents were removed because of highly unusual or nonsensical responses. While uncommon in panels such as AmeriSpeak, a small number of individuals may “spam” surveys (because they offer cash incentives) and thus provide nonusable data. This can be uncovered by looking for irregular response patterns. Two examples include a respondent that indicated spending thousands of dollars on food and transportation for visits in which they reported spending 5 days walking to the facility. Another respondent provided outlier costs for moving expenses while (purportedly) answering about their incarcerated grandson despite being in their early 30s.

Although they made it past the screener questions, some respondents indicated during the course of the survey that their family member’s incarceration ended pre-2016. Therefore, the final analytic sample of dyadic respondent-family member relationships was 1899, reported by 1472 individuals. The dyads serve as our primary unit of analysis throughout the paper—all expense estimates are for money exchanged between a given nonincarcerated adult and each incarcerated family member.

Survey protocol

The survey proceeded in three parts, the first two of which are used here. The first part asks respondents background questions about themselves and their incarcerated family member and the details of their family member’s incarceration. The second part takes respondents through a series of questions about common transactional costs between the respondent and each incarcerated family member that may arise during the incarceration period. We focus primarily on this second portion of the survey in the present analysis and describe it in detail below.

This series of questions asked respondents to provide information on the frequency and amount of costs incurred because of (1) collect phone calls with their incarcerated family member, which are borne by the nonincarcerated person; (2) contributions to commissary accounts (often referred to as putting money “on the books”) of their incarcerated family member; and (3) funds that were specifically earmarked for designated goods or services, such as essential goods (e.g., toiletries), health care, and classes/trainings. Collectively, we refer to these throughout the article as direct recurring expenses.

Because facilities differ widely in the way that commissary and monetary support is transacted, the survey offered flexibility and discretion in terms of the potential overlap between commissary and earmarked spending. Respondents were asked about what types of costs were covered under commissary and/or what their earmarked financial support was used on (e.g., clothing, food, and electronic tablets), although information at that level is not analyzed or used here.

In addition, respondents could report the frequency for each of the three types of expenses as follows: more than once per week, weekly, monthly, multiple times per year, yearly, and less than yearly. For our analyses, we convert all expenses to monthly to facilitate easier interpretation and comparison. Conversions are as follows if a respondent reported expenses: monthly, the quantity was unchanged; weekly or more than once per week, the quantity was multiplied by 4; yearly or multiple times per year, the quantity was divided by 12; less than yearly, the quantity was divided by 18.

This left us with the ability to analyze such direct recurring expenses equalized to a monthly amount. It also presents a conservative estimate of costs because individuals who have contributed more than once per week are multiplied by 4 and individuals who contributed multiple times per year are divided by 12. Furthermore, most expenses were reported monthly and weekly (which is naturally scalable to monthly), with more 50% of the dyad-level observations in each cost measure and about 60% in total. We describe our subsequent analytic approach below.

Measures

Monthly costs for collect calls, commissary, and direct goods were added together to get a “total” direct recurring expenses estimate. We provide descriptives of the total direct recurring expenses over (i) people who incurred any such expenses and (ii) all people, including those who did not pay for anything and have $0 in direct recurring expenses (thereby zero-inflating the measures). Results for each direct recurring expense are provided in the Supplementary Materials (figs. S1 to S3), which make it possible to see estimates and distributions for each expense type separately.

In addition to the reported direct recurring expenses, we also calculated the amount as a proportion of a respondent’s self-reported monthly household income. We used a question from the survey that asked the average annual household income during the time of the family member’s incarceration, in which respondents were given 18 different salary ranges from less than $5000 to $200,000 or more. Therefore, to get a denominator for the proportion of monthly income spent, we took the midpoint of the reported income range and divided it by 12 (except for the highest income level, for which $250,000 was used).

Estimation and analysis

To give a sense of the distribution of responses, we provide box plots that display the means (rather than medians, which are reported in Table 1), 25th and 75th percentiles (Q1 and Q3, respectively), whiskers [calculated as Q1 − 1.5 * interquartile range (IQR) and Q3 + 1.5 * IQR], and outlier points (values more extreme than the whiskers).

We took a number of steps to improve the stability and reliability of these estimates, in particular for the means, given that extreme outliers (which may be true outliers or products of misentry or misunderstanding of a survey question) will bias such estimates upward (48). Many different strategies are used for outlier detection and handling with limited consensus on best practices (49).

Therefore, we take a three-pronged approach throughout the process of analysis and interpretation that allows us to retain as much data as possible while limiting the influence of any plausible outliers that may not represent accurate responses. We note that these strategies are all inherently conservative, i.e., they bias any estimate toward zero given that all outliers are in the far-right tail of the distribution.

First, we rely more heavily on interpreting the medians throughout the paper, given their heightened resistance to outliers and skewed distributions. Medians best represent the true middle in this context, analogous to conventional practices of reporting the median of other financially related information (e.g., median household income, median housing prices, etc.). Means are only visualized in figures and are otherwise left as part of a secondary effort to understand distributional patterns in costs. In this context, as with the contexts above, the median individual in the survey represents what the average person in the target population would have contributed in direct recurring expenses to their family member per month. This is arguably a more meaningful quantity than the average amount individuals in the population contributed.

Second, we rely on the truncation of the most extreme values for costs. In the presentation of overall results (Table 1 and Figs. 1 to 3), we take the final expense estimate (after combining the different categories of commissary, collect calls, and designated giving) and subsequently truncate at the 99th percentile, i.e., we exclude any cost values larger than the 99th percentile value. Figure 4 shows the mean values at different truncation cutoffs. In figs. S1 to S3, which break down direct expenses by the specific category (commissary, collect calls, and designated giving), we truncate at the 99th percentile separately for each cost item.

Our choice of 99th percentile was driven by this being the best balance of reducing the effect of outliers—as demonstrated by the largest percentage change in cost from a one-percentile difference in truncation level (see Fig. 4)—yet also being the least invasive choice, given that it only cuts out 1% of responses. Said another way, it removes the least data while limiting the influence of particularly extreme outliers.

In a similar vein, for proportions of household income (as used in Fig. 2), we relied on top-coding. The truncated costs were used in the numerator to create the proportion of household income, but some proportions resulted in values greater than 1. Therefore, any proportions greater than 1 (i.e., when their indicated costs incurred exceeded their reported household income) were simply set to 1. We decided to top-code these values rather than truncate to retain information, because the numerator is already truncated and because proportions are naturally bounded at 1.

Third, we rely on bootstrapping to calculate robust confidence intervals around the mean without having to satisfy any parametric assumptions (50, 51). Bootstrap is a Monte Carlo resampling technique that provides flexibility and robust applicability to nearly any statistical estimation problem, especially in cases where distributional assumptions are not met. During bootstrapping, observations from the data are randomly sampled with replacement and the estimator of interest is calculated at each resample and then aggregated across the bootstraps, such as through averaging (52–54).

Since its inception, there has been much discussion and debate about how many bootstrap iterations are sufficient. Ultimately, this choice is dependent on the estimator and computational feasibility. It has been shown that even B as small as 50 will yield stable estimates (55), although the general rule of thumb suggests using at least B = 1000, which is often computationally manageable with modern computing (56).

Our setup requires the minimal computational burden of a well-defined estimator; therefore, we used B = 10,000 resamples to calculate 95% studentized bootstrap-t confidence intervals (which are overlaid on the means in the box plot figures). Under normality assumptions, a (1 − α)100% confidence interval of the sample mean $\overline{x}$ can be written as $(\overline{x}-z_{1-\frac{\mathrm{\alpha }}{2}}\raisebox{1ex}{$s$}\!\left/ \!\raisebox{-1ex}{$\sqrt{n}$}\right.,\overline{x}+z_{1-\mathrm{\alpha }/2}\raisebox{1ex}{$s$}\!\left/ \!\raisebox{-1ex}{$\sqrt{n}$}\right.)$ , where s is the sample standard deviation and $z_{1-\mathrm{\alpha }/2}$ is the standard normal value at the (1 − α/2)th quantile—called a z score (e.g., for a 95% confidence interval, this is 1.96). However, the bootstrap-t method yields confidence intervals that are second-order accurate and robust to violations of normality and skewness. The bootstrap interval looks similar to the normal interval defined above, with the exception that bootstrapped t-statistics will replace the z scores. The bootstrap-t has been advised as the best bootstrap technique for inferences about a mean (45).

To formalize the procedure used in our analysis, let X_i denote each observation of costs of the i = 1, …, n dyads in our sample, and let X_Bi* denote each resampled observation in bootstrap sample B of size n, where B = 1, …, b. For each bootstrap sample B, the mean, ${\overline{x}}_B=\frac{1}{n}{\displaystyle \sum _{k=1}^n}{X}_{B_k}$ , and variance, $s_B^2=\frac{1}{n-1}{\displaystyle \sum _{k=1}^n}{(X_{B_k}-{\overline{x}}_B)}^2$ , are computed. These are then used within each bootstrap sample to calculate a t-statistic: $t_B^{\ast }=\frac{{\overline{x}}_B-\overline{x}}{\sqrt{\raisebox{1ex}{$s_B^2$}\!\left/ \!\raisebox{-1ex}{$n$}\right.}}$ , thereby generating a vector of t-statistics of length B, which can be denoted as ${\mathbf{t}}_{\mathbf{B}}^{\mathbf{\ast }}=[t_1^{\ast },\dots ,t_B^{\ast }]$.

To get the t-statistics for the final bootstrapped interval, the vectors of t-statistics are ordered and the t-statistics corresponding to the lower and upper (α/2)th quantiles are taken as the multipliers for the lower and upper bounds, respectively. The final 95% bootstrap-t confidence interval for the mean is thus

$$[\overline{x}-t_{\left(0.025\right)B}^{\ast }\raisebox{1ex}{$s$}\!\left/ \!\raisebox{-1ex}{$\sqrt{n}$}\right.,\overline{x}+t_{\left(0.975\right)B}^{\ast }\raisebox{1ex}{$s$}\!\left/ \!\raisebox{-1ex}{$\sqrt{n}$}\right.]$$ (1)

Weighting

All the reported means and proportions are weighted estimates to provide generalizable results to the larger population. Weights provided by NORC were created at the respondent level. Given that our analysis is conducted at the dyad level—where a given respondent could have more than one response if they had multiple incarcerated family members—these original weights needed adjusting so that an individual respondent would not be “double-weighted” when providing information for two incarcerated family members.

Therefore, we made a simple adjustment such that for analyses at the dyad level, if a person responded for two family members, their respondent-level weight value was divided in half, thus creating a dyad-level weight. Weights for respondents who only responded for a single incarcerated family member were left unchanged. By simply dividing the original weights in half, the sum of the dyad-level weights still corresponded to the population totals that were used to calibrate the individual-level weights.

Robustness and analytic considerations

There are a number of challenges in asking about expenses incurred during a family member’s incarceration. One hurdle is that the policies and procedures vary widely across different jail and prison facilities in terms of how money is routed through commissary, the availability of collect calls (which are typically operated by independent private for-profit companies), scheduling of designated giving or providing packages, etc., and potential caps or upper limits on the amount that can be spent by a given family member or incarcerated person. Furthermore, there are very little existing rigorous estimates of these costs that can be used as benchmarks (particularly at the dyad level). It is also likely that some individuals who provided money to their incarcerated family members were ignorant to the details of how the money was being used or even what bucket(s) the contributions fell under in the first place.

We also recognized that the timing of one’s expenses likely varied on a case-by-case basis: Some may have provided regular financial support each week, while others could incur expenses more infrequently and sporadically. The fact that our estimates are not a point-in-time estimate also makes it difficult to account for any longitudinal changes in costs over time or the influence of inflation; if we anchor a given incarceration stint to the middle year of one’s incarceration stint (which leads to the loss of 30% of all observations because of missing entry and/or release year data), the median expense among those who contributed would be $203 in September 2023 dollars.

Therefore, we designed the survey in a way that maximized flexibility along a variety of dimensions. We asked respondents to report on the different items analyzed here separately and, in cases where there may have been overlap (e.g., they put money on their incarcerated family member’s commissary account but also provided additional money specifically for something like medical expenses or food), allowed them to report expenses independently of their previous responses. For each cost bucket, we also allowed respondents to choose the frequency and amount given at said frequency. While this results in a new challenge—that we have to multiply or divide these different amounts to arrive at a standardized monthly value—the alternative was forcing respondents to estimate their costs in a time frame that may not have been salient to them, which would present a range of (more serious, in our view) computational and interpretation issues.

Another challenge presented by the survey was handling outlier values. As described above, a handful of cases seemed to be obviously erroneous or faked data and were discarded. However, it was more difficult to arrive at a determination about other high number values. Some of these may have been “true” outliers in the sense that the respondent really did incur an exorbitant expense in a given category, or they may have been caused by a data entry error (e.g., typing in the wrong value, adding an extra “0” by accident, etc.) or by a conceptual misunderstanding (e.g., entering a summed yearly value after selecting monthly frequency). Instead of trying to arbitrate or discern the underlying nature of the outliers, we took a number of steps outlined above, especially truncation and bootstrapping, to preserve data while providing some adjustment to the most extreme values and bounding the mean values to offer insight into the level of uncertainty.

Closely related to the outlier issue is that of consistent (i.e., across all respondents) misreporting as a result of issues of memory, recall, or mental estimation. As discussed in the main text, we show in table S2 how our main parameters fluctuate—but not in a way that undermines our primary takeaways—under different assumptions of over- and underreporting for the entire sample.

There was a small proportion of missing data for the only three of our noncost measures: relationship (1.4%), the income measure used in the denominator for the proportion of income (1.7%), and the measure of whether the person was in jail or prison (5.9%). Values were left missing given the small proportion and minimal impact, and to retain the analysis on the pure responses of individuals.

The preceding limitations concentrate on issues that are primarily computational in nature. A final consideration, which relates more to macrolevel interpretation and conception, is that we analyze expenses at the dyad or relationship level but have little insight into (i) household-level or family dynamics of the survey respondent, (ii) the extent of financial support that a given incarcerated person is receiving from family members other than our respondent, and (iii) the level of financial capacity that the incarcerated person has to support themselves with their own prior earnings or savings.

Relatedly, some of the proportions of income that we calculated with the given information yielded values over 1, which imply respondents who are (reportedly) contributing more than their household income. Any values that were greater than 1 were then set equal to 1 (this was about 5% of calculated proportions). There could be a few reasons why this arises: genuine cases where someone goes into debt or draws from other external financial sources or some type of response issue such as data entry error, misreporting, or misremembering. Without any ability to determine the underlying reason behind such responses, we set this blanket rule because such responses are likely almost entirely a reporting error or extreme/outlier instances.

For instance, it may be that some respondents are reporting combined household-level expenses (e.g., shared between themselves and a spouse) that they incurred to support their incarcerated family member. It also may be the case that some respondents gave less than they otherwise would or could have because they knew that another family member, or the incarcerated person themselves, had sufficient financial means to provide ample support. While only an estimate, table S6 shows respondents’ best guess (at the dyad level) of what percentage their own expenditures were contributing to their incarcerated family member (relative to the entirety of financial support the incarcerated family member was receiving). Most respondents (about 63%) indicated that they were contributing less than a quarter of the total amount that their family member was receiving, which, along with Table 2, implies that the aggregate costs of incarceration are far greater than the direct expenses that we focus on here.

Supplementary Materials

The PDF file includes:

Figs. S1 to S3

Tables S1 to S6

AmeriSpeak panel and survey technical documentation

Legends for data S1 to S3

References

Other Supplementary Material for this manuscript includes the following:

Data S1 to S3