Disease outbreaks, healthcare utilization, and on-time immunization in the first year of life

Journal of Health Economics 67 (2019) 102212

Contents lists available at ScienceDirect

Journal of Health Economics journal homepage: www.elsevier.com/locate/econbase

Disease outbreaks, healthcare utilization, and on-time immunization in the first year of life夽 Jessamyn Schaller a,d,∗ , Lisa Schulkind b , Teny Shapiro c a

University of Arizona, Eller College of Management, Department of Economics, 1130 E Helen Street Suite 401, Tucson, AZ 85721-0108, United States University of North Carolina at Charlotte, Belk College of Business, Department of Economics, 9201 University City Blvd, Charlotte, NC 28223-0001, United States c Slack, Inc., 500 Howard Street, San Francisco, CA 94105, United States d Claremont McKenna College, Robert Day School of Economics and Finance, 500 East Ninth Street, Claremont, CA 91711, United States b

a r t i c l e

i n f o

Article history: Received 26 September 2017 Received in revised form 30 January 2019 Accepted 30 May 2019 Available online 6 July 2019 JEL classification: I12 I18 J13 Keywords: Immunizations Vaccinations Infant health Public health Disease Health care

a b s t r a c t This paper examines the determinants of parental decisions about infant immunization. Using the exact timing of vaccination relative to birth, we estimate the effects of local pertussis outbreaks occurring in utero and during the first two months of life on the likelihood of on-time initial immunization for pertussis and other diseases. We find that parents respond to changes in perceived disease risk: pertussis outbreaks within a state increase the rate of on-time receipt of the pertussis vaccine at two months of age, particularly among low-socioeconomic status (SES) subgroups. In addition, we find that pertussis outbreaks increase the likelihood of immunization against other vaccine-preventable diseases. Spillover effects in low-SES subgroups are as large as direct effects and are present only for vaccines given during the same visit as the pertussis vaccine, which suggests that provider contact may be a key factor in infant vaccination decisions in poor families. © 2019 Elsevier B.V. All rights reserved.

1. Introduction The United States experienced a dramatic decline in the incidence of vaccine-preventable diseases over the 20th century. This phenomenon, which resulted in substantial reductions in infant

夽 We are grateful to Kitt Carpenter, Emily Lawler, Emily Oster, and Chloe East for helpful suggestions and for sharing data with us. We would also like to thank seminar participants at the NBER Health Economics program meeting, the UC Davis Center for Poverty Research, UNC Greensboro, UNC Chapel Hill Public Policy, the Bureau of Economic Analysis, Lafayette College, Davidson College, Elon College and conference participants at the Western Economic Association, the Southern Economic Association, the Association for Public Policy Analysis and Management and the South Carolina Applied Micro Day for their helpful comments and suggestions. We also acknowledge financial support from a UNC Charlotte Faculty Research Grant and from the University of Arizona Center for Management Innovations in Healthcare. Restricted use NIS data files were created by Karon C. Lewis, at the National Center for Health Statistics Research Data Center. ∗ Corresponding author. E-mail addresses: [email protected] (J. Schaller), [email protected] (L. Schulkind), [email protected] (T. Shapiro). https://doi.org/10.1016/j.jhealeco.2019.05.009 0167-6296/© 2019 Elsevier B.V. All rights reserved.

and child morbidity and mortality, is attributable in large part to the success of a large scale public health campaign to promote universal immunization (Centers for Disease Control and Prevention, 1999). Though overall rates of vaccine coverage in the U.S. have been high in recent decades, disparities in immunization coverage have persisted across socioeconomic and racial groups. In particular, children who are poor, black, and from low-income neighborhoods, and those with younger, less-educated, and unmarried mothers are more likely to fall behind the recommended schedule for childhood vaccines (Feemster et al., 2009; Luman et al., 2005). In this paper, we examine the determinants of parental decisions about infant immunization, exploring the roles of perceived disease risk, attitudes about vaccination, and healthcare provider contact for vaccine takeup in the first year of life. We focus on the timing of a child’s first immunization doses for two reasons. First, the period encompassing pregnancy and early infancy is a time when many parents form opinions about vaccination and establish healthcare patterns for their children. Since most of the vaccines given to infants are part of a sequence, obtaining the first set of immunizations on time might help establish a routine with a care

2

J. Schaller et al. / Journal of Health Economics 67 (2019) 102212

provider and make it more likely that subsequent vaccines will be received on time as well. Second, vaccine delay is an important contributor to disparities in vaccination rates along socioeconomic and demographic dimensions. Many children, particularly those in disadvantaged families, will ultimately end up fully vaccinated by the time they start school but will experience substantial delays in their vaccination schedules during infancy and early childhood. As infants and toddlers have the highest likelihood of contracting vaccine-preventable diseases and of experiencing serious complications from them, vaccine delay has potentially-serious health implications (Feemster et al., 2009; Luman et al., 2005). We begin by examining whether local outbreaks of a vaccinepreventable illness affect the rate of on-time initiation of immunization for that illness at two months of age. Estimating the immunization response to outbreaks provides important insight into the role of perceived disease risk in influencing parents’ vaccination decisions. For some children, local outbreaks will raise the near-term risk of exposure to illness, increasing the health benefit from of on-time vaccination. Outbreaks also make disease risk more salient overall, possibly providing updated information to parents who previously underestimated the risk of illness (Oster, 2018). In order to explore whether information about disease risk affects infant immunization, we construct outbreak indicators using weekly state-level counts of pertussis (whooping cough) cases from the Center for Disease Control (CDC) and link them to child-level data from the National Immunization Survey. Using exact date of birth and the timing of each vaccine dose relative to birth date, we estimate the effects of pertussis outbreaks that occur in utero and during the first two months of life on the likelihood that the first dose of the diphtheria, tetanus, and pertussis (DTaP) vaccine is received within 75 days of birth. We find that parents respond to increases in perceived disease risk—a local pertussis outbreak occurring in utero or between birth and two months of age increases the probability that an infant’s first pertussis immunization occurs on-time. Splitting our sample along demographic and socioeconomic dimensions, we find that the increase in on-time vaccination at two months is entirely concentrated among children in low-income households, minority children, children with less-educated mothers, and children of unmarried mothers. We find no significant effects of outbreaks on the likelihood of on-time initiation of infant immunization among children in non-poor families, white children, or children with college-educated mothers. We confirm that these results are robust across different regression specifications and for a variety of outbreak definitions. In the second part of our study, we explore whether pertussis outbreaks affect the likelihood that infants receive vaccines for other diseases on time. Estimating the spillover effects of pertussis outbreaks allows us to gain insight into the role of information and the importance of healthcare provider contact in influencing parental decisions about immunization. First, we examine whether a pertussis outbreak increases on time receipt of other vaccines that are recommended to be given at the same age as the first DTaP dose (polio and Haemophilus influenzae type B (Hib)). We find that when a pertussis outbreak induces low-SES parents to vaccinate their infants against pertussis, their likelihood of obtaining the other vaccines offered at the same age increases by almost as much. As an outbreak of pertussis does not cause an increase in the likelihood of contracting other diseases, these effects are likely attributable either to changes in perceptions about vaccination and disease resulting from the outbreak or to reductions in the time cost or other healthcare access costs that result from seeking out the pertussis vaccine. In order to gain further insight into the mechanisms driving parental vaccination decisions, we next estimate the effect of pertussis outbreaks in utero on the likelihood that a child receives

immunizations that are typically recommended to be received at separate visits from the initial DTaP dose—immunizations at birth for hepatitis-B and immunizations for varicella, and measles, mumps, and rubella (MMR) around 12–15 months of age. Importantly, we find that pertussis outbreaks occurring in utero or in the first two months do not result in an increase in receipt of vaccinations for other disease at birth and at 12–15 months. This finding suggests that the increase in the two-month immunizations is not the result of broad changes in attitudes or information about vaccination or disease. In fact, for some groups, we find evidence of negative substitution effects: pertussis outbreaks lead to small decreases in the likelihood that these other vaccines are received during the recommended time frames. While we cannot separately identify all possible mechanisms behind our results, the nature of the spillover effects that we find—positive spillovers for same-visit vaccines and substitution for different-visit vaccines—and the fact that they are concentrated in disadvantaged groups with lower baseline use of medical care suggest that contact with healthcare providers may be an important factor contributing to lower on-time vaccination. Our hypothesis about the importance of provider contact is supported by reports from low-income focus-groups (Coker et al., 2009; Lannon et al., 1995), additional evidence of cross-visit substitution related to vaccination (Fiks et al., 2008), and the results from our own exploration of socioeconomic disparities in infant medical care use, access to quality medical care, and the characteristics of vaccination visits from the Medical Expenditure Panel Survey (MEPS).1 Our study complements recent studies of the immunization of older children by Oster (2018) and Carpenter and Lawler (2019). Oster studies the effects of pertussis outbreaks on the receipt of pertussis immunization among school-aged children using countylevel data from 12 US states, and also finds that outbreaks increase immunization coverage. She finds weak spillover effects of pertussis outbreaks on immunization for measles among school-aged children and interprets them as evidence for overall changes in parents’ perceptions of vaccination. Carpenter and Lawler (2019) study the effects of state laws requiring a tetanus, diphtheria, and pertussis (TDaP) booster shot prior to middle school enrollment on immunization rates of middle school students, and also find evidence of cross-vaccination spillovers. In particular, they find that TDaP mandates increased adolescent vaccination rates for meningococcal disease and human papillomavirus—vaccines that are also recommended for adolescents—with larger effects in lowSES subgroups. Though neither of the papers on older children focuses on provider contact as the key reason for spillovers, the findings in both papers are consistent with our results for infants and our interpretation of them. In both settings, the vaccinations that experience positive spillovers can be received at the same visit as the primary vaccination. The results from this paper have several important implications for health policy. First, we show that disease outbreaks affect the on-time initiation of infant immunization. This implies that dissemination of information about disease risk has the potential to increase rates of immunization coverage among very young infants—the population with the highest risk of serious complications from illness. Second, we find spillover effects of pertussis outbreaks on vaccination for other illnesses at the same visit, which implies that vaccine outreach programs and policies that help disadvantaged families to overcome the barriers associated with regular utilization of medical care in the first year of life may help to reduce the risk of vaccine delay. Finally, we find evidence of substitution across sets of vaccines that are received at different visits

1

Results from this analysis are presented in Table 8 and discussed in Section 7.

J. Schaller et al. / Journal of Health Economics 67 (2019) 102212

among some disadvantaged subgroups. That is, for black children and children with single mothers, the increase in the likelihood of obtaining the first set of vaccines on time at two months is accompanied by a reduction in the likelihood of obtaining a separate set of vaccines typically offered between 12 and 15 months. This finding suggests an important role for educating families about the importance of obtaining all of the recommended vaccines and doses and further highlights the importance of reducing barriers to consistent provider contact.

2. Healthcare access and socioeconomic disparities in immunization Because increasing immunization coverage has been a central goal of United States health policy for several decades, there exists a substantial literature, most of which is in the fields of public health and epidemiology, exploring patterns in U.S. immunization rates in the cross-section and over time. Many studies have documented cross-sectional patterns in vaccination coverage that mirror patterns of socioeconomic inequality. In particular, black and Hispanic children, children from low-income families, and children with younger, less-educated, and unmarried mothers, are less likely to adhere to the vaccine schedule recommended by the Center for Disease Control (CDC) than the general population (Chu et al., 2004; Feemster et al., 2009; Guttmann et al., 2006; Luman et al., 2003; Smith et al., 2004). Discrepancies in vaccine coverage by socioeconomic status are especially pronounced when vaccine delay is taken into account, with substantial fractions of children in disadvantaged groups spending many months of their early childhood underimmunized (Luman et al., 2005). Existing research highlights the connection between immunization patterns and overall healthcare utilization patterns in disadvantaged groups. Descriptive evidence suggests that lower levels of healthcare provider contact may prevent compliance with the recommended vaccine schedule among low-SES children. In particular, studies have found that disadvantaged children are substantially less likely to maintain a regular schedule of wellchild visits than more-advantaged children (Ronsaville and Hakim, 2000), and that children with low continuity in their healthcare provision are less likely to comply with the recommended vaccine schedule (Luman et al., 2005). Provider quality may also be important, as studies have found that under-immunization is correlated with the receipt of care from a public health clinic or a less-experienced pediatric-care provider (Feemster et al., 2009; Guttmann et al., 2006). Evidence suggests that access and convenience may be particularly important factors contributing to low levels of provider contact and reduced compliance with immunization guidelines, particularly among single-parent or two-earner families. In particular, research has found that children with lower spatial accessibility to pediatricians are less likely to be vaccinated on time (Fu et al., 2009; LeBaron et al., 2001), and two focus group studies highlight scheduling challenges and transportation costs as key barriers to immunization facing disadvantaged families (Lannon et al., 1995; Coker et al., 2009). Another study finds evidence of cross-visit substitution generated by vaccine receipt: children who receive immunizations at sick-child visits are less-likely to attend subsequent well-child visits (Fiks et al., 2008). Though it does not include children, a recent field experiment provides particularly convincing evidence on the importance of convenience and the opportunity cost associated with obtaining immunizations: assigning working adults to receive subsidized vaccination at their workplace during the workweek rather than over the weekend increased vaccine take-up by 112% compared with workers who were offered vaccines on a weekend day (Hoffmann et al., 2019).

3

The roll of financial cost in driving socioeconomic disparities in immunization is unclear. Quasi-experimental evidence suggests that increases in access to immunization coverage resulting from public insurance expansions and private insurance mandates increases compliance with immunization recommendations (Joyce and Racine, 2005; Chang, 2016). However, in the cross-section, the groups with the lowest average out-of-pocket costs have the lowest average rates of immunization coverage. In fact, all children covered by public health insurance (Medicaid and the State Children’s Health Insurance Plan) have access to immunizations and preventative care with no out of pocket cost, and the federal Vaccines for Children (VFC) program and Section 317 of the Public Health Service Act additionally ensure that vaccines are provided at no cost to children who are uninsured and children whose insurance plans do not cover vaccines. There are, however, gaps in coverage and barriers to takeup of these benefits. In particular, the VFC program does allow some providers to charge administration fees to children who are not covered by public insurance and families need to locate a VFC provider or public health department to obtain subsidized vaccines. As a result, while children in low-SES families have lower out-of-pocket costs for vaccines on average, some may end up paying out-of-pocket fees and those fees may be a burden in low-income families (Molinari et al., 2007). Because of the documented disparities in on-time vaccination, we stratify all of our analysis in this paper along socioeconomic and demographic dimensions. In particular, we separately consider the effects of outbreaks on the likelihood of on-time vaccination by poverty status, race, maternal education, and parents’ marital status. The particular emphasis in the literature on the importance of healthcare access and provider contact motivates our analysis of the spillover effects of pertussis outbreaks to vaccination for other illness and separation of the effects on same-visit and different-visit vaccine doses. Later, in Section 7), we briefly revisit the discussion about reasons for disparities in immunization by socioeconomic status and supplement our NIS analysis with statistics from the MEPS on healthcare utilization, parents’ perceived access to care and quality of care, and characteristics and costs of vaccine visits in the first year of life for the SES groups we consider. 3. Conceptual framework A simple conceptual framework, similar to that outlined by Oster (2018), aids in identifying the many mechanisms that might drive changes in immunization in response to disease outbreaks. In this framework, parents make vaccine decisions by weighing perceived vaccine costs against perceived vaccine benefits. The perceived benefit from vaccination for a particular disease depends on the perceived severity of the disease and the perceived excess probability of contracting the disease in the absence of a vaccine, which is in turn a function of perceived likelihood of exposure to illness and perceived vaccine efficacy. Perceived costs of vaccination include the expected health costs of vaccinating as well as any healthcare access costs associated with vaccination. The expected health costs of vaccination include injection pain and a small risk of allergic reaction or other medical complications from the vaccine. For some families, they also include other concerns about vaccine safety such as the purported link between vaccines and autism (Omer et al., 2009).2 Access costs include the costs of obtaining information about vaccine recommendations, locating a provider,

2 The word perception is important here: survey responses reveal large differences in perceptions about disease susceptibility, disease severity, vaccine efficacy, and vaccine safety between the parents of vaccinated kids and parents who have taken nonmedical exemptions from vaccine from school immunization requirements (Salmon et al., 2005).

4

J. Schaller et al. / Journal of Health Economics 67 (2019) 102212

and scheduling an appointment, as well as copayments or fees, transportation costs, and time costs. We can use this conceptual framework to generate predictions about the direct and spillover effects of disease outbreaks on immunization. Importantly, we assume that a local pertussis outbreak (the external shock in our analysis) causes the perceived risk of exposure to pertussis to increase, increasing the perceived benefit from pertussis immunization. For this assumption to hold, local disease outbreaks must be adequately publicized within a state. We assume that parents find out about outbreaks through the local news media, from employers, from schools, or from medical providers. Oster (2018) provides supporting evidence on this front by documenting increases in both internet searches and local news articles related to pertussis following an outbreak. We also confirm that internet search activity increases during outbreaks defined in our data (see Appendix Table 1). By increasing the perceived benefits from pertussis immunization, we expect that outbreaks should increase on-time immunization for pertussis. However, other effects may be present as well. For example, positive effects of changes in perceived disease risk may be enhanced or counteracted by new information about vaccinations that parents are exposed to during an outbreak. Healthcare providers and news articles may be more likely to discuss the risks and benefits of vaccination during an outbreak. It is also possible that for some families the cost of immunization may change as a result of the outbreak if public health departments or medical providers make vaccines more readily available during an outbreak or if there is a temporary shortage of vaccines. Within a cost–benefit framework, especially if information matters, there are many reasons to expect heterogeneity across population subgroups in the immunization response to outbreaks. For one thing, some parents may be more responsive to disease outbreaks simply because they are closer to the margin to begin with: the excess of their perceived vaccine cost over their perceived benefit is smaller before the outbreak. One reason for this might be that certain groups perceive additional costs of vaccines. Along these lines, previous researchers have highlighted disparities in reported reasons for vaccination by socioeconomic status—while low-income parents are likely to delay vaccination because of reasons related to healthcare access, more-advantaged parents are more likely to intentionally choose not to vaccinate because of perceptions about vaccines and personal preferences (for example, concern about the purported relationship between vaccines and autism or an ideological opposition to government mandates). Omer et al. (2008) summarize this discussion and cite related studies. It is also possible that outbreaks may generate larger changes in the perceived benefits and costs of vaccination among some groups. On the benefit side, some groups may be more likely to receive information about an outbreak. These could be parents who have more exposure to local news, parents who have older children in school, or parents who have more frequent contact with medical providers. Local health departments and medical providers also might enact information outreach programs that target specific groups or geographic areas. Some groups may also be more likely to be exposed to an outbreak, perhaps because of geographic proximity or lifestyle risk factors. Finally, it is possible that some parents update their expected risk of illness more when an outbreak occurs because they were underestimating it more to begin with—possibly parents with less health knowledge or less frequent contact with medical care providers. On the cost side, changes in vaccine availability during an outbreak could differentially affect certain groups or local areas. Turning to possible spillover effects, a pertussis outbreak may also change the perceived benefits and costs of vaccinating for other diseases. For example, awareness of a pertussis outbreak could

change parents’ perceptions about the overall risk of contagious illness or could lead to changes in general attitudes about vaccination. There are also direct spillover effects from going to the doctor to get a pertussis vaccine. Additional provider contact, particularly in early infancy when parents may not have much experience with vaccination, may expose parents to health information that changes their long-term vaccination plans. A trip to the doctor for a pertussis vaccine also eliminates the opportunity and access costs associated with obtaining other vaccines—costs of locating and scheduling with a provider, the time cost of the visit, and the copay for the visit. Unlike spillover effects resulting from changes in information, the latter spillover effects should be seen only for vaccines given at the same visit as the pertussis vaccine. We would also expect that these spillover effects should only be seen for families who had a positive direct effect from the outbreak and families that are constrained in their healthcare utilization—those who are not already maintaining a regular schedule of well-child visits. 4. Data In order to measure the effect of disease outbreaks on immunization behavior, we combine two datasets—one that contains individual level information on immunizations and another that contains information on the number of pertussis cases in each state and week.3 4.1. Outbreak data The outbreak data come from the Centers for Disease Control and Prevention (CDC)’s Morbidity and Mortality Weekly Reports (MMWR). Each week, the MMWR reports provide provisional counts of selected notifiable diseases for each state. The reports also include revised counts for the prior year. Using those weekly reports, we compiled a database that covers the years 1996–2012 and includes both the provisional counts of pertussis cases for each state and week and the revised counts. In order to examine whether or not parents respond to local outbreaks of a vaccine-preventable disease, we first need to define what an “outbreak” is. Our primary measure of an outbreak is an indicator for whether the number of weekly pertussis cases is above the 95th or 99th percentile of disease counts within a state over time, where percentile thresholds are calculated using non-zero weekly counts. Our choice to focus on 95th and 99th percentile outbreaks is informed by an analysis of internet search data from Google Trends. As shown in Table 1, we find that in weeks where the count is above either the 95th or the 99th percentile, search activity for the term “whooping cough” significantly increases within a state.4 Although we also find increases in search activity with lower thresholds, the increases are largest at these higher thresholds, and the coefficients for being between the 95th and 98th percentile or above the 99th percentile are statistically different from the coefficients for the lower categories. Appendix B.3 describes the details of this analysis. In both the Google Trends analysis and in our main analysis, we use a moving average to smooth out any lumpiness in reporting. Inspection of the data suggests that states may occasionally skip

3 We have also collected disease count data for measles, mumps and varicella (Chicken Pox), but pertussis is the only one that is available throughout the entire time period. Infrequently reported diseases (<1000 cases reported in the previous year) are not reported by state, so our state level measure is not available for all diseases in all years. For this reason, and because there is significantly more variation in exposure to pertussis outbreaks, we focus on pertussis. 4 Search activity is measured with an index between 0 and 100, provided by Google Trends, with higher numbers indicating more search activity relative to other weeks for that state during the sample period.

J. Schaller et al. / Journal of Health Economics 67 (2019) 102212 Table 1 Google trends searches for “whooping cough”.

Any cases 50–75th Percentile 75–90th Percentile 90–95th Percentile 95–99th Percentile 99–100th Percentile

5

Table 2 Summary statistics. (1) Search activity

Variable

Observations

Mean

SD

−0.527 (0.615) −0.100 (0.627) 1.567* (0.800) 2.209* (1.157) 4.901*** (1.795) 9.364*** (3.090)

2-Month immunizations DTaP by 75 days Hib by 75 days Polio by 75 days First DTaP, age in days First Hib, age in days First polio, age in days

164,018 164,018 164,018 161,128 160,935 160,294

0.788 0.775 0.769 76.087 77.376 76.557

0.409 0.417 0.421 57.778 59.666 55.145

Other immunizations Hepatitis B by 6 days MCV by 470 days Varicella by 470 days # DTaP by 19 months UTD on DTaP by 19 months

164,018 164,018 164,018 164,018 164,018

0.563 0.745 0.728 3.577 0.693

0.496 0.436 0.445 0.791 0.461

Demographic variables Black Hispanic Poor HS degree College degree First born Male

164,018 164,018 164,018 164,018 164,018 164,018 164,018

0.143 0.286 0.311 0.507 0.307 0.427 0.514

0.351 0.452 0.463 0.500 0.461 0.495 0.500

Notes: The key explanatory variables are indicators for the level of pertussis cases. Each variable is equal to one if the average of the current week and the three previous weeks falls within the percentile range listed in the first column, for that state. The omitted category is zero cases. The regression includes state and week × year fixed effects. Standard errors (in parentheses) are clustered at the state level. * p < 0.10. *** p < 0.01

reporting in a week, and effectively report multiple weeks at once. Using the four week average, which encompasses the current week as well as the previous three weeks, helps us avoid accidentally counting those weeks as being above the threshold.5 Fig. 1 gives an example of an outbreak based on our preferred definition. The top left panel displays the 4 week moving average of the number of cases, by week, in Idaho, and the horizontal line gives Idaho’s 99th percentile threshold for pertussis cases (27.75). During 1997, Idaho had a number of weeks above its 99th percentile. At the same time, we can see that in three nearby states, there weren’t any weeks above the threshold. Appendix B includes additional descriptive statistics related to the outbreaks and patterns of outbreaks across time and space. Table B1 gives the 95th and 99th percentile thresholds for each state and Fig. B.1 shows the distribution of outbreaks across the weeks in our sample period. 4.2. Immunization data The immunization data come from the National Immunization Survey (NIS). The NIS is sponsored by the CDC, and it is used to generate official estimates of vaccine coverage at the state and national level. The study collects information on randomly selected households with children between the ages of 19 months and 35 months via a household telephone survey and a mail survey of vaccination providers. The data are available from 1995 through 20126 and include information on household demographics, as well as the types of vaccinations received and the child’s age in days at vaccine receipt.7 We use the restricted use version of the NIS, which includes each child’s state of residence and exact date of birth. This allows us to create a variable that indicates whether or not a pertussis outbreak occurred in a child’s state while they were in utero or during their first two months of life. Our primary outcome of interest is an indicator for whether or not an individual received their first DTaP immunization on time. The recommended age is two months (61 days), and we allow for a two week grace period to account for appointment scheduling conflicts, for a total of 75 days. We also examine whether an increase in on-time receipt of the

5 There are some additional data related issues that require additional cleaning of the data. The procedure we use in our main specification, as well as two alternative procedures that we consider are detailed in Appendix B. 6 Data for 2004 is not available in the restricted use version of the NIS dataset. 7 We only include children who have “adequate provider information” on vaccine receipt, as defined by the NIS, and drop a small fraction of children with missing information for any of our individual control variables.

Notes: This table gives summary statistics for the National Immunization Survey (NIS). The sample for this table is limited to the sample of children who are included in our main set of DTaP regressions, as described in Section 5. Means are weighted using survey weights.

DTaP vaccination spills over to other two-month immunizations, which can be received during the same visit, as well as a number of immunizations that are recommended at different visits. Table 2 displays the summary statistics for the immunization and demographic variables in our NIS estimation sample. Approximately 79% of children receive their first DTaP immunization by 75 days. The percentage of children receiving their other 2-month immunizations on time is slightly lower, but similar (Hib, Polio). There is a long right tail in the distribution, however, with an average age of first DTaP of 76 days, but a standard deviation of 58.8 The percentage of children receiving other immunizations on time is lower than for the 2-month immunizations. We see that only 56% of children receive their first Hepatitis B immunization at birth (by 6 days), 75% of children receive their first measles containing vaccine by 15 months (470 days), and 72% of children receive their varicella vaccination by 15 months. On average, children have received 3.6 doses of DTaP by 19 months, with only 68% having received the full recommended sequence of four doses by 19 months. 5. Estimation strategy We use a fixed effects model with state-specific linear time trends to estimate the effect of disease outbreaks on the propensity for children to be immunized at different points in time. The model includes month of birth and state fixed effects so that our identification relies on variation in the timing of outbreaks across and within states. In other words, we estimate whether birth cohorts that experience a very high number of pertussis cases relative to other birth cohorts in the same state, are more likely to be immunized on time, controlling for aggregate (nationwide) variation in immunization across birth-cohorts and linear trends in immunization within state.

8 The number of observations for the “Age in Days” variables is lower than the number of observations for the “by 75 days” variables, because children who have not received the vaccination have a missing value for age in days. They are coded not having had the immunization by 75 days.

6

J. Schaller et al. / Journal of Health Economics 67 (2019) 102212

Fig. 1. Pertussis Cases in 1997. Notes: This displays the 4 week moving average of the number of pertussis cases per week in a sample of four states. The horizontal line gives the 99th percentile for the 4 week moving average of weekly cases in each state. Data come from the Centers for Disease Control and Prevention’s Morbidity and Mortality Weekly Reports.

Our preferred estimation specification is: DTP75isw = ˛ + ˇOutbreaksw + ωXisw + Ssw + s + m + s + isw (1) where DTP75isw is an indicator for whether or not individual i, living in state s, and born in week w received their first DTaP immunization by the time they are 75 days old. Outbreaksw is a variable that indicates exposure to a disease outbreak during the one-year period spanning the ten months prior to birth and the two months after birth, for a child born in week w and in state s. For our main results, we show the effects of both 95th percentile and 99th percentile outbreaks, where the outbreak cutoffs are defined relative to own-state pertussis counts. In the remaining tables, we limit our attention to 99th percentile outbreaks in the interest of space. The results are similar when we use 95th percentile instead. Xisw is a vector of individual child characteristics that includes birth order, child gender, child race and ethnicity, maternal education, and family poverty status. Ssw is a vector of state policy variables that measure vaccine related mandates as well as Medicaid and CHIP thresholds at the time of birth. These variables and their sources are described in Appendix C.  s and  m are fixed effects for state and month-by-year of birth and  s are state specific linear time trends. Standard errors are clustered by state and the regressions are weighted using NIS survey weights. ˇ is the regression coefficient of interest, representing the effect of an outbreak in utero or early infancy on the probability of on-time immunization at two months of age. 6. Main results 6.1. Effect of pertussis outbreaks on DTaP by 75 days Our first question is whether information about disease risk directly affects the timing of infant vaccination for the disease in question. In Table 3, we show the estimated effects of experiencing a pertussis outbreak during the past 12 months on the likelihood that an infant receives his or her first DTaP vaccine dose on time by 75 days of age. We estimate our model for the full NIS sample and also stratify by poverty status, race/ethnicity, maternal education, and parental marital status.

Each column in Table 3 presents the results of estimating Eq. (1) separately for the group listed at the top of the column (including all control variables except for the stratifying variable). In this table, we show results both for 99th percentile outbreaks and 95th percentile outbreaks. Before examining our regression results, we first verify that the summary statistics for on-time receipt of the first DTaP dose match the patterns that have been established in the literature. The row titled “Y Mean” near the bottom of Table 3 provides the weighted mean for the subgroup listed at the top of the column. For example, the means at the bottom of the second and third columns tell us that 82.2% of children above the poverty line receive their first DTaP immunization by the time they are 75 days old, while only 71.3% of children below the poverty line do. Across the remaining columns, the discrepancy in on-time immunization rates between high- and low-SES families is apparent for each set of subgroups: Black and Hispanic children, children of less-educated mothers, and children of unmarried mothers all have lower rates of immunization. The regression coefficients in Table 3 show that there is a positive immunization response to disease outbreaks in low-SES subgroups. We find that children in families below the poverty line are 2.2–3.0 percentage points more likely to obtain their first dose of DTaP on time if they experience a pertussis outbreak, depending on the severity of the outbreak. These values represent a 7.7–10.5% reduction in the size of non-vaccinating population at that age (note that the number displayed in italics below each coefficient’s standard error is the coefficient expressed a percentage of the portion of that group that is not vaccinated on time). We find similarlylarge effects among black and Hispanic children, with estimates by race/ethnicity ranging from 8% to 16% of the non-vaccinating population in each subgroup. The estimated treatment effect is also large for children whose mothers have less than a high school education—a 12–13.4% effect, though the results are less precisely estimated due to smaller sample size and are not statistically significant. Finally, the coefficients suggest that children of single mothers are more likely to get vaccinated on time after an outbreak than children of married mothers, though those results are only significant for 99th percentile outbreaks. Notably, in the results shown in Table 3, we find no evidence of any effects of pertussis outbreaks on the likelihood of obtaining the first pertussis vaccine on time among the non-poor, white, and college-educated groups.

J. Schaller et al. / Journal of Health Economics 67 (2019) 102212

7

Table 3 Effects of pertussis outbreaks on DTaP immunization by 75 days. Full

Poverty status

Race and ethnicity

Sample

Not poor

Poor

White

Black

Hispanic

LTHS

HS grad

College

Not married

Married

−0.000 (0.006) 0.0000

0.030** (0.014) 0.1045

−0.004 (0.006) −0.0213

0.029* (0.015) 0.1090

0.037** (0.016) 0.1595

0.039 (0.024) 0.1336

0.008 (0.008) 0.0351

−0.004 (0.006) −0.0294

0.026* (0.015) 0.0949

0.005 (0.006) 0.0273

Panel A: 99th percentile cutoff 0.011 Pertussis outbreak (0.008) (12 months) 0.0519

Mother’s education

Marital status

Panel B: 95th percentile cutoff Pertussis outbreak (12 months)

0.009 (0.006) 0.0425

0.002 (0.004) 0.0112

0.022* (0.013) 0.0767

0.001 (0.006) 0.0053

0.035** (0.016) 0.1316

0.019* (0.011) 0.0819

0.035 (0.021) 0.1199

0.002 (0.006) 0.0088

0.004 (0.005) 0.0294

0.016 (0.011) 0.0584

0.007 (0.005) 0.0383

Y mean Observations

0.788 164,018

0.822 128,401

0.713 35,617

0.812 110,098

0.734 19,594

0.768 34,326

0.708 18,275

0.772 73,293

0.864 72,450

0.726 40,157

0.817 123,861

Notes: All specifications include the individual and state policy controls described in Section 5, as well as state linear time trends, state fixed effects, and month-year of birth fixed effects. The numbers in italics represent the treatment effects as a fraction of the unvaccinated population. The key explanatory variable is an indicator for having experienced an outbreak of pertussis in the state of birth during the period starting 10 months prior to birth and ending 2 months after birth. This indicator is equal to one if the family experienced a four week period in which the total number of pertussis cases in the state exceeded the 99th or 95th percentile of all 4-week case totals for that state during the sample period. Standard errors (in parentheses) are clustered at the state level. Observations are weighted using NIS sampling weights. * p < 0.10. ** p < 0.05. Table 4 Effects of pertussis outbreaks on DTaP immunization by 75 days, specification comparison, 99th percentile. (1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Panel A: Full sample Pertussis outbreak (12 months)

0.011* (0.006)

0.014** (0.006)

0.013** (0.006)

0.010 (0.007)

0.011 (0.008)

0.011 (0.007)

0.012 (0.007)

0.011 (0.008)

Y mean Observations

0.788 164,018

0.788 164,018

0.788 164,018

0.788 164,018

0.788 164,018

0.788 164,018

0.788 164,018

0.788 164,018

Pertussis outbreak (12 months)

0.037*** (0.011)

0.035*** (0.010)

0.036*** (0.011)

0.031** (0.012)

0.029** (0.013)

0.035*** (0.011)

0.033** (0.013)

0.030** (0.014)

Y mean Observations

0.713 35,617

0.713 35,617

0.713 35,617

0.713 35,617

0.713 35,617

0.713 35,617

0.713 35,617

0.713 35,617

Individual control State fixed effects Year fixed effects State trends Month-year fixed effects State policy controls

No No No No No No

Yes No No No No No

Yes Yes No No No No

Yes Yes Yes No No No

Yes Yes Yes Yes No No

Yes Yes No No Yes No

Yes Yes No Yes Yes No

Yes Yes No Yes Yes Yes

Panel B: Below the poverty line

Notes: The key explanatory variable is an indicator for having experienced an outbreak of pertussis in the state of birth during the period starting 10 months prior to birth and ending 2 months after birth. This variable is equal to one if the family experienced a four week period in which the total number of pertussis cases in the state exceeded the 99th percentile of all 4-week case totals for that state during the sample period. Individual control variables include dummies for child gender, race/ethnicity, and poverty status, and mother’s education and marital status. Standard errors (in parentheses) are clustered at the state level. Observations are weighted using NIS sampling weights. * p < 0.10. ** p < 0.05. *** p < 0.01.

We explore the robustness of the estimated effects of outbreaks in Table 4, checking the sensitivity of the coefficients to the choice of added controls for the full NIS sample and for the sample of children below the poverty line. In the interest of space, we focus our attention in this and all remaining tables on 99th percentile outbreaks. We begin by estimating the association between experiencing an outbreak and likelihood of immunization in the pooled NIS sample with no additional controls, and then add sets of controls and fixed effects one at a time. The coefficient is remarkably stable across specifications. We conservatively select the specification in column (8), which includes month-year fixed effects, state-specific time trends, and controls for state vaccine policy parameters, as our preferred specification. Results based on alternative specifications and for 95th percentile outbreaks are similar (and often more precisely estimated) than the results presented in this paper and are available upon request.

Next, we provide support for our identification strategy with a falsification test. If within-state differences across birth-week cohorts in the likelihood of experiencing a pertussis outbreak are associated with other unobservable changes (in attitudes about vaccination, for example) that are related to the likelihood of ontime vaccination, our results will be biased. There are two reasons why this is unlikely to be a significant concern here. First, the inclusion of state-specific linear trends should account for secular trends in vaccination preferences and immunization choices. Second, the bias is most likely to be negative—cohorts with less inclination to vaccinate should be more likely to experience outbreaks—so it is unlikely to explain the significant positive coefficients that we find in Table 3. Nonetheless, we conduct a falsification test in which we include outbreaks that occur in the 12 months after the twomonth immunizations. These results, presented in Appendix Table A1, show no significant association between receiving the first

8

J. Schaller et al. / Journal of Health Economics 67 (2019) 102212

DTaP dose on time and subsequent outbreaks. We found that these results are similar for alternate time periods (outbreaks between 3 and 6 months) and with additional controls for earlier outbreaks. We also examine the sensitivity of our main results to alternative ways of defining pertussis outbreaks. We have chosen to use state-specific outbreak thresholds in our main specification because public narratives surrounding pertussis outbreaks typically refer to case counts rather than population shares and compare those counts to area-specific histories. However, using state-specific thresholds may cause outbreaks to be identified where there were none, if some states have relatively steady levels of cases with no outlying periods. As an alternative way to define outbreaks, we divide 4-week case counts by state population and create an indicator variable that is equal to one if the rate of pertussis is higher than the 99th or 95th percentile of disease rates for all 4-week periods in all states in the sample. Since this alternative outbreak indicator is based on population shares rather than case counts it may be more strongly correlated with the true disease risk that individuals face. However, defining outbreaks this way may also lead to a concentration of outbreaks in certain states that have higher disease rates in general and may cause us to miss outbreaks occurring in states with lower overall per-capita risk. The results for national threshold, presented in Appendix Table A2 show generally the same pattern as our main results—disease outbreaks lead to increases in on-time vaccination among disadvantaged groups. Another potential concern is that states may not be the appropriate level of aggregation. This may be the case if people are only aware of or concerned about outbreaks that occur in their metro area or county, or if different groups are differentially likely to be exposed to state-level outbreaks. In order to address this, we use annual county level pertussis counts.9 The drawback of these data is that, because they are only available at the annual level, they do not allow us to identify the timing of outbreaks relative to the week of birth as precisely as the state data. We create a predicted outbreak indicator at the county level (described in detail in Appendix Section B) based on the state-level weekly counts and the annual shares of counts within a state that are attributable to each county in that state. First, we use the county-level data to explore whether state outbreaks are experienced disproportionately by low-SES groups in our study. In fact, as shown in Fig. B.2, we find that county-years in which a county’s share of state pertussis cases is large relative to the county’s share of state-population occur in counties with higher per capita income, lower poverty rates, and lower percent black and Hispanic. Regression results showing the estimated effects of pertussis outbreaks on pertussis immunization, using this county-level indicator and county-level fixed effects are presented in Table A3 and reveal patterns that are again consistent with our main results. Finally, we ran a specification that includes multiple outbreak percentiles together, including separate indicators for highest 4week period in the 50–74th percentile, 75–89th percentile, and 90–94th percentile along with the 95th and 99th percentile indicators. The results from this specification are presented in Appendix Table A4. Though the estimates are noisy, they show that the effects of outbreaks on the likelihood of on-time immunization in the poor, black, and single-parent samples are generally increasing with outbreak severity and are the strongest for 99th percentile outbreaks. 6.2. Spillover effects of pertussis outbreaks Next, we turn our attention to the spillover effects of pertussis outbreaks onto immunization for other illnesses. We first estimate the effects of experiencing a pertussis outbreak in the previous year

9 The county level case counts were provided to us by Emily Oster, who obtained them from the CDC.

on the likelihood that children receive vaccinations for polio and Haemophilus influenzae type b (Hib)—two immunizations that are also recommended at two months of age—by 75 days. These results are presented in Table 5. We find that a pertussis outbreak results in significant increases in the likelihood that infants are immunized on-time for polio and Hib, with effects concentrated among the same low-SES groups that had the largest direct effects. The results are only slightly smaller in magnitude than the direct effects shown in Table 3. For example, among poor children, the coefficient represents a 9.9% reduction in the non-vaccination rate for polio and an 8.0% reduction in the non-vaccination rate for Hib. These findings imply that once children in low-SES groups visit a vaccine provider to obtain a pertussis shot, they obtain other shots that are offered at the same visit.10 As discussed above, spillover effects of pertussis outbreaks to other vaccines could reflect changes in information and attitudes about vaccines that result from an outbreak. For vaccines given at the same visit as the pertussis vaccine, spillover effects could also reflect reductions in the marginal cost of obtaining the vaccines and/or additional information about other immunizations shared by the provider during the visit. In particular, visiting the doctor for a pertussis vaccine removes costs such as scheduling, transportation, and visit co-pays that would otherwise be associated with vaccinating for other illnesses and the additional provider contact provides an opportunity for the provider to share information about the existence, benefits and recommended schedule of other vaccinations. This information might be limited to vaccinations that can be provided during the current visit, or it could be more comprehensive in nature. In order to distinguish between these possible mechanisms driving the spillover effects in Table 5, we next turn our attention to vaccines that are typically received separately from the first DTaP dose: the first dose of hepatitis-B, which is recommended in the hospital at birth, and the first doses of MMR and varicella, both of which are recommended at 12–15 months. We expect that the reduction in marginal costs will be reflected in spillovers only to immunizations received at the same visit and only for the groups that responded to the outbreaks with an increased likelihood of on-time DTaP immunization. If the extra information received during the visit includes information about future immunizations, it could affect same visit immunizations and the timelines of later vaccinations. If the spillovers are driven by general changes in information and attitudes about vaccines that result from an outbreak, we would expect to see an increase in both different and same visit immunizations. Receipt of the first Hepatitis-B dose at birth is the cleanest test of the potential alternative mechanisms described above, since the CDC recommendation about when this dose should be given is unambiguous and it is recommended within the same general timeframe as our key outcome variables (early infancy). With MMR and varicella, the potential mechanisms are more complicated and interpretation is a bit more challenging. For these two vaccines, the CDC recommendation about when they should be received is ambiguous—the first dose is recommended between 12 months and 15 months. They are typically given at 12 months, when there is no DTaP dose given. However, if a child does not obtain them at 12

10 It is important to note that in recent years, more options for combination vaccines have become available. It could be that we observe spillover effects simply because the DTaP immunization is given in the same injection as the others. If this is the case, it might change how we think about the “decision” to get additional vaccinations at the same visit. Before 2009, there was a combination available that bundled the polio vaccination with DTaP, but no combination that included both DTaP and Hib. We check whether there are any differences in spillovers to polio vs. Hib pre-2009, before Hib was available as a combination. We find no difference, suggesting that the observed spillovers are not being driven by combination shots. These results are available upon request.

J. Schaller et al. / Journal of Health Economics 67 (2019) 102212

9

Table 5 Spillover effects on other immunizations received at the same visit as the first DTaP dose, 99th percentile outbreaks. Full

Poverty status

Race and ethnicity

Sample

Not poor

Poor

White

Black

Hispanic

Mother’s education LTHS

HS grad

College

Marital status Not married

Married

Panel A: Polio by 75 days 0.010 Pertussis outbreak (0.007) −10m to 2m 0.0433

−0.001 (0.005) −0.0050

0.029** (0.014) 0.0986

−0.005 (0.005) −0.0234

0.026* (0.013) 0.0949

0.036** (0.016) 0.1481

0.035 (0.023) 0.1178

0.008 (0.007) 0.0329

−0.004 (0.006) −0.0234

0.024 (0.015) 0.0848

0.004 (0.005) 0.0193

Y mean Observations

0.769 164,018

0.798 128,401

0.706 35,617

0.786 110,098

0.726 19,594

0.757 34,326

0.703 18,275

0.757 73,293

0.829 72,450

0.717 40,157

0.793 123,861

Pertussis outbreak −10m to 2m

0.011* (0.006) 0.0489

0.003 (0.005) 0.0158

0.024* (0.014) 0.0795

−0.003 (0.006) −0.0151

0.021 (0.013) 0.0745

0.039*** (0.013) 0.1579

0.034 (0.021) 0.1115

0.011 (0.007) 0.0456

−0.004 (0.006) −0.0268

0.023* (0.013) 0.0799

0.006 (0.005) 0.0306

Y mean Observations

0.775 164,018

0.810 128,401

0.698 35,617

0.801 110,098

0.718 19,594

0.753 34,326

0.695 18,275

0.759 73,293

0.851 72,450

0.712 40,157

0.804 123,861

Panel B: Hib by 75 days

Notes: All specifications include the individual and state policy controls described in Section 5, as well as state linear time trends, state fixed effects, and month-year of birth fixed effects. The numbers in italics represent the treatment effects as a fraction of the unvaccinated population. The key explanatory variables are indicators for having experienced an outbreak of pertussis in the state of birth during the relevant reference period. These indicators are equal to one if the family experienced a four week period in which the total number of pertussis cases in the state exceeded the 99th percentile of all 4-week case totals for that state during the sample period. Standard errors (in parentheses) are clustered at the state level. Observations are weighted using NIS sampling weights. * p < 0.10. ** p < 0.05. *** p < 0.01.

months, they can be given 15 months, which is when the 4th DTaP dose is typically given. Here, our outcome variables are indicators for receiving each shot by 15 months of age, or by the end of the CDC recommendation window. The expected effects of a pertussis outbreak that occurs in utero or in the first two months of life on these later immunizations will incorporate changes in vaccination attitudes overall that result from the outbreak, changes in the likelihood of attending the 12 and 15 month visits, and changes in the likelihood of getting the immunizations at each visit. Estimates of the spillover effects from pertussis outbreaks to the hepatitis-B, MMR, and varicella vaccines are presented in Table 6. In Panel A, which shows the effects on hepatitis-B, the outbreak indicator covers the 12 months prior to birth, as the outcome in question occurs at the time of birth. In Panels B and C, which show the effects on 15 month immunizations, we control for later outbreaks as well, including an indicator for outbreaks occurring between 3 and 15 months of life. Across all three panels, we find no evidence that early-life pertussis outbreaks generate the same positive spillover effects that we saw in the previous table for vaccines that are recommended to be received separately from DTaP doses. In fact, we find the opposite: pertussis outbreaks seem to generate substitution across immunizations. In particular, black children and children of unmarried mothers are substantially less likely to obtain both MMR and varicella on time if they experienced a pertussis outbreak in utero or in early life. Taken together, our results suggest that low-SES groups are constrained with respect to their health care utilization, and that this constraint is affecting infant immunization coverage in these groups. Among poor, less-educated, minority, and unmarried groups, a pertussis outbreak increases the likelihood of getting the first set of vaccines on time—not just pertussis, but all vaccines that are recommended at two months of age. However, for some families, these increases are accompanied by a reduction in the likelihood that later vaccines are received on-time. It appears that even if the information shared by the provider at the 2-month visit includes information about future immunizations, it is not enough to overcome other barriers. Though we can only speculate about the reasons for this, the fact that the substitution effects are particularly strong for the children of unmarried mothers and for the children of black mothers, who are substantially more likely to be unmar-

ried and also more likely to work than mothers in any of the other groups, suggests that it may be driven by families experiencing time constraints. 6.3. Long-run immunization effects Finally, we explore whether on-time initiation of pertussis immunization has an effect on the number of pertussis doses that are received and whether children are ultimately up-to-date on their pertussis sequence. Table 7 shows the effect of a pertussis outbreak during the 12 month periods beginning 6 months before birth and at 6 months old on immunizations received by 19 months (we shift the timing of the outbreak shocks by a few months in order to better cover all outbreaks experienced in infancy). The results in Panel A show that an outbreak during either of those time periods increases the number of DTaP vaccinations received by 19 months, though only very slightly. By that time, the full 4 dose sequence should be completed. The mean doses for each subgroups range between 3.4 and 3.7, and all but one of the coefficients translates to less than 1% of the relevant subgroup mean. Interestingly, the increases in the total number of vaccine doses is seen not only in the low-SES subgroups but also in the more advantaged groups. Along with a few positive cross-vaccine spillovers seen for pertussis outbreaks occurring between 3 and 15 months of age in Table 6, these results suggest that non-poor, white, and college-educated families do respond somewhat to outbreaks, but do so by adjusting the timing of later vaccine doses rather than by accelerating the timing of the first dose. Panel B displays the results for an indicator for whether a child has received all four recommended DTaP immunizations by 19 months of age. Here, we see that while a pertussis outbreak during the period six months before to six months after birth has a positive, and often statistically significant, effect on being up-to-date, a later outbreak does not. This suggests that an earlier outbreak, which causes an on-time initiation of the sequence also encourages parents to make sure their children are fully up to date by 19 months. However once parents delay, they are unable to complete the sequence on time, even if a later outbreak encourages them to initiate the sequence earlier than they would have in the absence of an outbreak.

10

J. Schaller et al. / Journal of Health Economics 67 (2019) 102212

Table 6 Spillover effects on immunizations received at separate visits from the first DTaP Dose, 99th percentile outbreaks. Full

Poverty status

Race and ethnicity

Sample

Not poor

Poor

White

Black

Hispanic

LTHS

HS grad

College

Not married

Married

Panel A: Hepatitis B at birth 0.001 Pertussis outbreak (0.006) −12m to 0m 0.0023

0.005 (0.007) 0.0111

−0.003 (0.011) −0.0079

−0.003 (0.007) −0.0068

−0.024 (0.016) −0.0612

0.015 (0.011) 0.0361

0.010 (0.014) 0.0251

0.001 (0.008) 0.0024

0.001 (0.008) 0.0021

0.001 (0.009) 0.0026

0.002 (0.007) 0.0044

Y mean Observations

0.550 124,283

0.619 34,570

0.556 106,787

0.608 18,773

0.584 33,293

0.602 17,689

0.585 70,772

0.531 70,392

0.617 38,820

0.550 120,033

0.572 158,853

Mother’s education

Marital status

Panel B: Measles containing vaccine by 15 months Pertussis outbreak −10m to 2m

−0.010* (0.006) −0.0394

−0.007 (0.006) −0.0280

−0.015 (0.011) −0.0570

−0.004 (0.007) −0.0149

−0.037* (0.021) −0.1326

−0.010 (0.012) −0.0476

−0.009 (0.013) −0.0347

−0.015 (0.009) −0.0568

−0.002 (0.006) −0.0085

−0.031** (0.012) −0.1140

−0.001 (0.006) −0.0041

Pertussis outbreak 3m to 15m

0.010 (0.007) 0.0394

0.012 (0.009) 0.0480

0.002 (0.011) 0.0076

0.018* (0.010) 0.0669

0.002 (0.016) 0.0072

−0.000 (0.011) 0.0000

0.018 (0.014) 0.0695

0.005 (0.009) 0.0189

0.013 (0.011) 0.0556

0.014 (0.013) 0.0515

0.005 (0.006) 0.0204

Y mean Observations

0.746 152,814

0.750 119,975

0.737 32,839

0.731 102,720

0.721 18,250

0.790 31,844

0.741 16,988

0.736 68,170

0.766 67,656

0.728 37,157

0.755 115,657

Panel C: Varicella by 15 months Pertussis outbreak −10m to 2m

−0.012** (0.005) −0.0444

−0.011* (0.006) −0.0423

−0.015 (0.010) −0.0514

−0.003 (0.006) −0.0108

−0.053** (0.025) −0.1840

−0.019 (0.015) −0.0779

−0.004 (0.013) −0.0137

−0.015 (0.010) −0.0530

−0.012 (0.009) −0.0511

−0.032** (0.012) −0.1107

−0.003 (0.005) −0.0115

Pertussis outbreak 3m to 15m

0.008 (0.006) 0.0296

0.010 (0.008) 0.0385

0.005 (0.010) 0.0171

0.022 (0.013) 0.0791

0.003 (0.016) 0.0104

−0.014 (0.013) −0.0574

0.016 (0.018) 0.0546

0.002 (0.009) 0.0071

0.015** (0.007) 0.0638

0.002 (0.012) 0.0069

0.009 (0.006) 0.0345

Y mean Observations

0.730 152,814

0.740 119,975

0.708 32,839

0.722 102,720

0.712 18,250

0.756 31,844

0.707 16,988

0.717 68,170

0.765 67,656

0.711 37,157

0.739 115,657

Notes: All specifications include the individual and state policy controls described in Section 5, as well as state linear time trends, state fixed effects, and month-year of birth fixed effects. The numbers in italics represent the treatment effects as a fraction of the unvaccinated population. The key explanatory variables are indicators for having experienced an outbreak of pertussis in the state of birth during the relevant reference period. These indicators are equal to one if the family experienced a four week period in which the total number of pertussis cases in the state exceeded the 99th percentile of all 4-week case totals for that state during the sample period. Standard errors (in parentheses) are clustered at the state level. Observations are weighted using NIS sampling weights. * p < 0.10. ** p < 0.05.

7. Discussion Though many mechanisms are possible, our results suggest that contact with healthcare providers may be an important factor in the underimmunization of children in low-SES families. As discussed in Section 2, this narrative is supported by evidence from crosssectional and qualitative studies of disadvantaged populations. To aid in the interpretation of our results, we conduct our own brief analysis of differences in healthcare utilization, cost, and perceived access and quality by family income and race using parental reports about children ages 12 to 18 months in seventeen waves of the Medical Expenditure Panel Survey (MEPS). The results are presented in Table 8. Consistent with previous evidence in the public health and medical literature, we observe large differences in the frequency of medical care utilization during the previous 12 months across columns—only 28% of children below the poverty line have attended five or more medical visits in the past 12 months, compared with 43% of children above the poverty line, and black children are far less likely to have attended five or more visits than white children. Poor and minority children are also more likely to have had no health visits at all in the past 12 months. Looking more closely at the distribution of visit counts over the past year (not shown), we see that, in fact, more than half of black and Hispanic children have had fewer than 3 medical visits in the past 12 months, while the mode response for white children is 5 or more. The large disparities in care utilization exist despite similar rates of overall insurance coverage across SES groups—approximately 92% of chil-

dren are covered by insurance, regardless of poverty status, and in fact, black children are more likely to be covered by insurance than white children. We also compare parental reports about access to healthcare and the quality of the care that their infants receive. We find that poor and minority parents are more likely to report difficulty with scheduling health appointments. They are also more likely to report concerns about the quality of care that they receive. For example, when asked how often their child’s doctor explained things in a way that they (the parent) could understand or how often a child’s healthcare provider listened to them, poor parents are more likely to respond “never” than non-poor parents. The MEPS also includes information about every interaction with medical providers that children have, including the reason for the visit, the location of the visit, and the out-of-pocket cost, which we average across visits for each child and then average across children. Patterns in the reported characteristics of vaccine visits confirm that poor children and black children are less likely to receive their vaccinations as part of a regular schedule of well child visits. In particular, vaccines are less likely to be received in an office setting (compared with an outpatient setting or an emergency room) and less likely to be received at a checkup. Parents in poor and minority groups are also more likely to report that vaccination was the primary reason for the visit rather than a checkup or a health diagnosis. Likely as a result of higher public insurance coverage, poor families pay lower out-of-pocket costs for vaccine visits, averaging only $2.41 per visit compared with $12.23 per visit.

J. Schaller et al. / Journal of Health Economics 67 (2019) 102212

11

Table 7 Effects of pertussis outbreaks on DTaP immunization coverage by 19 months, 99th percentile. Full

Poverty status

Race and ethnicity

Sample

Not poor

Poor

White

Black

Hispanic

LTHS

HS grad

College

Not married

Married

−0.012 (0.027) −0.0034

0.017* (0.009) 0.0047

−0.036 (0.040) −0.0103

0.015 (0.018) 0.0042

−0.023 (0.031) −0.0066

0.012 (0.014) 0.0034

0.026** (0.010) 0.0070

0.021 (0.024) 0.0060

0.002 (0.010) 0.0006

Panel A: Number of DTaP vaccinations 0.008 0.018* Pertussis outbreak (0.009) (0.012) −6m to 6m 0.0022 0.0050

Mother’s EDUCATION

Marital status

Pertussis outbreak 6m to 18m

0.015** (0.008) 0.0042

0.015** (0.007) 0.0041

0.024 (0.017) 0.0069

0.006 (0.009) 0.0017

0.007 (0.025) 0.0020

0.022 (0.015) 0.0062

0.023 (0.023) 0.0066

−0.001 (0.009) −0.0003

0.035*** (0.010) 0.0095

0.015 (0.017) 0.0043

0.013 (0.011) 0.0036

Y mean Observations

3.582 148,551

3.624 116,847

3.487 31,704

3.608 99,840

3.491 17,754

3.575 30,957

3.474 16,510

3.554 66,260

3.695 65,781

3.494 35,965

3.622 112,586

Panel B: Up to date on DTaP vaccinations Pertussis outbreak −6m to 6m

0.009 (0.006) 0.0294

0.010* (0.005) 0.0356

0.007 (0.017) 0.0193

0.005 (0.006) 0.0175

0.008 (0.019) 0.0217

0.020* (0.011) 0.0635

0.011 (0.020) 0.0303

0.004 (0.008) 0.0121

0.017*** (0.005) 0.0733

0.011 (0.014) 0.0302

0.008* (0.004) 0.0286

Pertussis outbreak 6m to 18m

0.002 (0.005) 0.0065

0.003 (0.005) 0.0107

0.003 (0.008) 0.0083

−0.007 (0.006) −0.0245

0.012 (0.013) 0.0326

0.011 (0.009) 0.0349

−0.001 (0.013) −0.0028

−0.003 (0.006) −0.0091

0.011* (0.007) 0.0474

−0.003 (0.011) −0.0082

0.003 (0.008) 0.0107

Y mean Observations

0.694 148,551

0.719 116,847

0.638 31,704

0.714 99,840

0.632 17,754

0.685 30,957

0.637 16,510

0.670 66,260

0.768 65,781

0.636 35,965

0.720 112,586

Notes: All specifications include the individual and state policy controls described in Section 5, as well as state linear time trends, state fixed effects, and month-year of birth fixed effects. In Panel A, the numbers in italics represent the treatment effects as a percent of the mean. In Panel B, the numbers in italics represent the treatment effects as a fraction of the unvaccinated population. The key explanatory variables are indicators for having experienced an outbreak of pertussis in the state of birth during the relevant reference period. These indicators are equal to one if the family experienced a four week period in which the total number of pertussis cases in the state exceeded the 99th percentile of all 4-week case totals for that state during the sample period. Standard errors (in parentheses) are clustered at the state level. Observations are weighted using NIS sampling weights. * p < 0.10. ** p < 0.05. *** p < 0.01. Table 8 Healthcare use and access by socioeconomic status: ages 12–18 months (MEPS). Full

Poverty status

Race and ethnicity

Sample

Not poor

Poor

White

Black

Hispanic

Healthcare use No appointments past 12 months 5 or more appointments past 12 months

0.06 0.40

0.05 0.43

0.09 0.28

0.05 0.48

0.07 0.26

0.10 0.30

Insurance coverage Medicaid Private insurance Covered by insurance

0.41 0.52 0.92

0.26 0.66 0.92

0.86 0.08 0.93

0.27 0.66 0.93

0.66 0.30 0.95

0.62 0.27 0.90

Healthcare access and quality Always can schedule appointments Never can schedule appointments Doc always listens Doc never listens Doc always explains Doc never explains

0.76 0.06 0.77 0.05 0.79 0.04

0.76 0.05 0.77 0.04 0.80 0.03

0.73 0.09 0.76 0.06 0.76 0.07

0.80 0.04 0.79 0.04 0.81 0.03

0.75 0.08 0.81 0.05 0.82 0.05

0.70 0.09 0.74 0.06 0.76 0.06

Vaccine visits Vaccines received in Office Vaccines received at checkup Vaccine reason for visit Vaccine visit out of pocket cost

0.98 0.54 0.42 8.99

0.98 0.58 0.39 12.23

0.97 0.46 0.49 2.41

0.99 0.63 0.34 12.20

0.96 0.49 0.48 5.05

0.97 0.46 0.49 6.33

Observations

12,365

4537

7828

3729

2095

4763

Notes: The data are from the 2000 to 2016 panels of the Medical Expenditure Panel Survey. The sample is limited to observations for children who are between 12 and 18 months old and is not weighted. Information on healthcare use, insurance coverage, and healthcare access and quality are from the full-year consolidated files, including the (parental) self-administered questionnaire. Information on vaccine visits is from the office-based, outpatient, and emergency visit files. Vaccine visits include all healthcare visits at which a vaccination was received.

The picture of health care utilization that emerges from the MEPS data is that, despite having similar rates of insurance coverage and lower out-of-pocket visit costs, low-SES kids are less likely to receive regular preventative healthcare than kids from more-

advantaged families and perceive lower access to information and quality of care from their providers. They are also less likely to obtain vaccines at checkups or diagnostic visits. Though the patterns in Table 8 could in part reflect reverse causality (lower desire

12

J. Schaller et al. / Journal of Health Economics 67 (2019) 102212

to vaccinate resulting in lower visit frequency among low-SES families), when taken with our main results, which show that low-SES families accelerate vaccination in response to disease risk, are willing to get all recommended vaccines when they do visit the doctor, and substitute visits across time periods, the statistics from the MEPS are further suggestive evidence that it is low provider contact, rather than other factors, that contributes to under-vaccination in low-SES families. 8. Conclusions Improving early childhood vaccination rates is an important policy goal to ensure the health of American children. A growing number of children are under-vaccinated, either for economic or personal reasons. The goal of this study has been to understand how changes in perceived disease risk and access costs alter parents’ decisions to initiate vaccination for their infants. By assessing the effects of pertussis outbreaks during pregnancy and early infancy on the procurement of vaccines that protect against pertussis, as well as those that protect against different diseases, we are able to explore the roles of perceptions about disease and vaccines and healthcare provider contact in the decision to vaccinate. For children below the poverty line, we find that a pertussis outbreak while a child is in utero increases the likelihood of obtaining the first pertussis vaccination on-time by three percentage points—the equivalent of a 10.5% reduction in the non-vaccinated population. Similarly, responses are largest for children whose mothers have less than a high school education, black and Hispanic children and those whose mothers are unmarried. Since low-income, lesseducated, and minority families are very responsive to changes in their perceived disease risk, these populations might be effectively targeted with policy efforts that improve information about the benefits of vaccination. We also find that when parents respond to a pertussis outbreak by getting their child immunized against pertussis, they often bundle that immunization with others that can be given at the same time. However, for some groups, this increase in immunization is accompanied by a decrease in the likelihood of getting other immunizations that are typically recommended to be received at separate visits from the first DTaP dose. These patterns reveal the significant role that provider contact plays in influencing parental decisions about immunization, particularly for the populations who are likely to be time- and resource-constrained. They are perhaps not surprising, as full immunization requires frequent visits to the doctor. In fact, perfect compliance with the CDC’s recommended vaccination schedule for infants and toddlers requires as many as six separate interactions with medical providers—at birth, two months, four months, six months, 12 months, and 15 months. These visits can involve monetary costs, time costs, and transportation costs. Policies that help eliminate these costs such as communitybased health centers or free neighborhood vaccination clinics can help facilitate vaccine initiation and follow-up for this population. The literature on improving attendance at prenatal visits may provide insight about how to facilitate well-child visits for the same population.

Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at https://doi.org/10.1016/j.jhealeco.2019.05. 009. References Carpenter, C.S., Lawler, E.C., 2019. Direct and spillover effects of middle school vaccination requirements. American Economic Journal: Economic Policy 11 (1), 95–125. Centers for Disease Control and Prevention, 1999. Achievements in public health, 1990–1999: control of infectious diseases. Morbid. Mortal. Wkly. Rep. 48 (29), 621–629. Chang, L.V., 2016. The effect of state insurance mandates on infant immunization rates. Health Econ. 25 (3), 372–386. Chu, S.Y., Barker, L.E., Smith, P.J., 2004. Racial/ethnic disparities in preschool immunizations: United states, 1996–2001. Am. J. Public Health 94 (6), 973–977. Coker, T.R., Chung, P.J., Cowgill, B.O., Chen, L., Rodriguez, M.A., 2009. Low-income parents’ views on the redesign of well-child care. Pediatrics 124 (1), 194–204. Feemster, K.A., Spain, C.V., Eberhart, M., Pati, S., Watson, B., 2009. Identifying infants at increased risk for late initiation of immunizations: maternal and provider characteristics. Public Health Rep. 124, 42–53. Fiks, A.G., Hunter, K.F., Localio, A.R., Grundmeier, R.W., Alessandrini, E.A., 2008. Impact of immunization at sick visits on well-child care. Pediatrics 121 (5), 898–905. Fu, L.Y., Cowan, N., McLaren, R., Engstrom, R., Teach, S.J., 2009. Spatial accessibility to providers and vaccination compliance among children with medicaid. Pediatrics 124 (6). Guttmann, A., Manuel, D., Dick, P.T., To, T., Lam, K., Stukel, T.A., 2006. Volume matters: physician practice characteristics and immunization coverage among young children insured through a universal health plan. Pediatrics 117 (3), 595–602. Hoffmann, M., Mosquera, R., Chadi, A., 2019. Vaccines at Work. Technical Report. Joyce, T., Racine, A., 2005. Chip shots: association between the state children’s health insurance programs and immunization rates. Pediatrics 115 (5), e526–e534. Lannon, C., Brack, V., Stuart, J., Caplow, M., McNeill, A., Bordley, W.C., Margolis, P., 1995. What mothers say about why poor children fall behind on immunizations: a summary of focus groups in North Carolina. Arch. Pediatr. Adolesc. Med. 149 (10), 1070–1075. LeBaron, C.W., Massoudi, M., Stevenson, J., Lyons, B., 2001. Vaccination coverage and physician distribution in the United States, 1997. Pediatrics 107 (3), e31. Luman, E.T., Barker, L.E., Shaw, K.M., McCauley, M.M., Buehler, J.W., Pickering, L.K., 2005. Timeliness of childhood vaccinations in the United States. J. Am. Med. Assoc. 293 (10), 1204–1211. Luman, E.T., McCauley, M.M., Shefer, A., Chu, S.Y., 2003. Maternal characteristics associated with vaccination of young children. Pediatrics 111 (Supplement 1), 1215–1218. Molinari, N.-A.M., Kolasa, M., Messonnier, M.L., Schieber, R.A., 2007. Out-of-pocket costs of childhood immunizations: a comparison by type of insurance plan. Pediatrics 120 (5). Omer, S.B., Enger, K.S., Moulton, L.H., Halsey, N.A., Stokley, S., Salmon, D.A., 2008. Geographic clustering of nonmedical exemptions to school immunization requirements and associations with geographic clustering of pertussis. Am. J. Epidemiol. 168 (12), 1389–1396. Omer, S.B., Salmon, D.A., Orenstein, W.A., deHart, P., Halsey, N., 2009. Vaccine refusal, mandatory immunization, and the risk of vaccine-preventable diseases. N. Engl. J. Med. 360 (19). Oster, E., 2018. Does disease cause vaccination? Disease outbreaks and vaccination response. J. Health Econ. 57, 90–101. Ronsaville, D.S., Hakim, R.B., 2000. Well child care in the United States: racial differences in compliance with guidelines. Am. J. Public Health 90 (9), 1436. Salmon, D.A., Moulton, L.H., Omer, S.B., Patricia deHart, M., Stokley, S., Halsey, N.A., 2005. Factors associated with refusal of childhood vaccines among parents of school-aged children: a case–control study. Arch. Pediatr. Adolesc. Med. 159 (5), 470–476. Smith, P.J., Chu, S.Y., Barker, L.E., 2004. Children who have received no vaccines: who are they and where do they live? Pediatrics 114 (1), 187–195.