Cold summer weather, constrained restoration, and very low birth weight in Sweden

Health & Place 22 (2013) 68–74

Contents lists available at SciVerse ScienceDirect

Health & Place journal homepage: www.elsevier.com/locate/healthplace

Cold summer weather, constrained restoration, and very low birth weight in Sweden Terry Hartig a,n, Ralph Catalano b a b

Institute for Housing and Urban Research, Uppsala University, Box 514, SE-75330 Uppsala, Sweden School of Public Health, University of California, Berkeley, CA 94720-7360, USA

art ic l e i nf o

a b s t r a c t

Article history: Received 20 October 2012 Received in revised form 13 March 2013 Accepted 17 March 2013 Available online 26 March 2013

In higher latitudes, relatively cold summer weather may constrain outdoor activities that provide relief from chronic stress. Chronic stress can affect human birth outcomes, including the length of gestation and so the birth weight of the infant. We tested the hypothesis that, in Sweden, the odds of very low birth weight (VLBW; o1500 g) vary inversely with mean monthly temperature for the summer months. We applied time-series modeling methods to nationally aggregated data on singleton births during the 456 months from January, 1973, through December, 2010. We found elevated odds of VLBW among male infants for relatively cold June and August temperatures. Unpleasant weather may figure in stress-related health outcomes, not only as a stressor, but also as a constraint on restoration. & 2013 Elsevier Ltd. All rights reserved.

Keywords: Ambient temperature Global climate change Physical activity Psychological restoration Psychological stress

1. Introduction Concerns about adverse effects of weather on human health have increased as evidence of global climate change has mounted (McMichael et al., 2006). Concern has centered on the temperature extremes, violent storms, and heavy rains that can quickly injure or kill people who lack suitable shelter. Yet, even when people can protect themselves from its extreme forms, weather may affect health through its effects on behavior. In the present paper we focus on the constraint of psychological restoration as one behavioral pathway through which weather can affect health. We consider how constrained restoration can engender very low birthweight, an outcome with long-lasting and far-reaching implications for the individuals involved and for the society in which they live (Gäddlin, 2011). The definition of constrained restoration refers to the general causes of chronic stress. Stress arises when environmental demands tax or exceed a person's adaptive resources (Cohen et al., 1997). It becomes chronic in part because of persistent demands and in part because of a persistent lack of adaptive resources. The lack of resources can follow from an inability to acquire new resources, to more effectively apply those one has, or to restore those depleted in efforts to cope. We focus here on the

n

Corresponding author. Tel.: +46 18 471 6532; fax: +46 18 471 6501. E-mail addresses: [email protected] (T. Hartig), [email protected] (R. Catalano). 1353-8292/$ - see front matter & 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.healthplace.2013.03.008

last of these reasons for resource insufficiency. If a person cannot adequately restore depleted resources because of prevailing environmental conditions, then the situation has one of the defining characteristics of constrained restoration. Another defining characteristic involves a distinction between conditions that constrain restoration and those that evoke stress. People would not ordinarily appraise the constraining condition as a stressor, but in a context dedicated to restoration they would nonetheless evaluate it negatively. We argue that weather can indirectly affect health by discouraging participation in outdoor activities that support restoration. If weather constrains restorative activities over an extended period, then chronic stress may go unrelieved. Chronic stress can cause mental and physical health to suffer in a variety of ways (Lovallo, 2005). In the present study, we consider the possibility that, at higher latitudes, relatively cold summer weather can lead to adverse birth outcomes by denying pregnant women sufficient relief from chronic stress. Using time-series analytic methods with nationally aggregated data for Sweden, we estimate the association between mean monthly temperature and the odds of very low birth weight for each of the summer months, looking across the years 1973–2010. In performing this test, we build upon a study which found that the dispensation of antidepressants in Sweden varied inversely with mean monthly temperatures for July across the years 1991 through 1998 (Hartig et al., 2007). We see several advantages in using Swedish data to study the constraint of restoration due to poor weather. Sweden has the dark and difficult winters characteristic of higher latitudes, and Swedes

T. Hartig, R. Catalano / Health & Place 22 (2013) 68–74

value the summer as a season for outdoor leisure activities. The evolution of Swedish vacation legislation reflects this positive regard for the summer. Statements about the superiority of a summer vacation for recreation and restoration appear in legislative proposals (e.g., Kunglig Majestäts propositionen no. 286, 1938, as cited in Andra Lagutskottet (1953) and Kunglig Majestät (1951)), inquiries commissioned by the government (Statens Offentliga Utredningar, 1944, 1967, 1975), and reports from legislative committees (Andra Lagutskottet, 1951) that span several decades. Swedish vacation legislation currently enables employees to take four consecutive weeks of paid vacation during the period June 1-August 31 (Ericson and Gustaffson, 1977). It thus ensures widespread access to restorative activities during the months in focus here, although in terms of the time needed and not the environmental conditions that some of those activities may require. In line with the vacation law, vacationing in Sweden shows strong seasonality, with a pronounced spike during the summer months. In keeping with the preferences acknowledged in the vacation legislation, population surveys show that Swedes generally do engage in more outdoor activity during the summer months than during cooler months (Statens Offentliga Utredningar, 1964; Statistiska Centralbyrån, 2004; cf. Matthews et al., 2001; Pivarnik et al., 2003). Similarly, observations of different outdoor spaces in one major city, Gothenburg, found that the number of people present varied with temperature across the year, and that temperature had a stronger association with the number of people present than did either the amount of cloud cover or wind speed (Eliasson et al., 2007). Presence outdoors also varies with the weather within the summer months. For example, during afternoons in one July, Thorsson et al. (2004) observed that the number of people resting in a large park in Gothenburg varied strongly with temperature. Despite the preference for warm weather during the summer, we do not regard relatively cold summer temperatures as stressful, given the readily available means to maintain thermal comfort (e.g., stay indoors, wear more clothing, and drink hot coffee). Rather than a stressful exposure, we view cold summer temperature as a condition that hinders activities that help people to recover from efforts to deal with role obligations and other stressful demands. Some demands may persist continuously over many months, and so underlie stress that has become chronic prior to the opportunities for restoration that open during the summer months. As a dependent variable, very low birth weight (VLBW) offers a crucial advantage for our study, in that much previous research has linked it with chronic stress (Hobel et al., 2008; Wadhwa et al., 2002). Research has pointed to several sources of persistent demands that may affect birth weight, including low socioeconomic position, racial discrimination, residence in an insecure neighborhood, and stressful work (Hobel and Culhane, 2003; Hogue et al., 2001; Katz, 2012). These circumstances can affect birth weight through increasingly well understood physiological mechanisms (Wadhwa et al., 2001). The elevated corticosteroid levels that attend chronic stress may make the “clock” for delivery run faster, leading to a preterm delivery (Hobel and Culhane, 2003). Elevated corticosteroid levels may also suppress immune function, allowing latent infections to become active and thereby increasing the risk of preterm delivery (Coussons-Read et al., 2007). Aside from preterm delivery, chronic stress may affect birth weight by restricting intrauterine growth (Hobel et al., 2008); however, more than 90% of very low weight infants come preterm (Crouse and Cassady, 1994). Like other people, pregnant women might suffer from constraints on restorative outdoor activities due to cold summer weather. Advice on vigorous physical activity during pregnancy has varied over the period under study because of differing opinions about the potential for harm to the fetus. The activity

69

of interest here however may impose only modest demands. As for other people, activities such as walking can offer restorative benefits to pregnant women (Da Costa et al., 2003), especially when done in pleasant outdoor surroundings (Hartig, 2007). Even women who must cope with the discomfort, functional limitations and other demands of late-stage pregnancy (Otchet et al., 1999) may benefit from access to the outdoors under agreeable weather conditions. Simply sitting in an outdoor space, especially one with a natural character such as a park, may offer restorative advantages over staying inside; the pregnant woman can passively enjoy the surroundings and gain psychological distance from stressful demands in the home or workplace (Hartig, 2007; Donovan et al., 2011). To the extent that their choices of activities and locations depend upon weather conditions, pregnant women may lose access to environments with relatively high restorative quality in times of cold summer weather. Another feature of pregnancy that supports our selection of dependent variable involves the possibility of escaping cold weather. Particularly in the third trimester, many women want to remain close to familiar medical facilities and expertise. Late-term pregnant women also face restrictions on air travel that might otherwise bring them to satisfactory places with relatively greater speed and comfort than possible with other modes of transportation (American College of Obstetricians and Gynecologists, 2001). For these reasons, even pregnant women of high socioeconomic status may not escape activity limiting weather late in pregnancy. Finally, relatives and friends who have planned to provide support as a woman nears the end of her pregnancy also have their mobility constrained. For some of them, providing support may mean taking on more domestic work. These conditions may hold during the time they take for vacation and also during that period of the summer during which, though not formally on vacation, they want to enjoy outdoor activities, with or without the pregnant woman. Similarly, should the woman already have children, she might have to contend with stressful consequences of their activity restrictions. Such household circumstances may feed stress contagion among those involved (Rook et al., 1991). In sum, a very low weight birth may follow chronic stress that has persisted because cold summer temperatures have constrained access to restorative outdoor environments by a pregnant woman, her family and her friends. Accordingly, we test the hypothesis that the odds of VLBW in Sweden have a negative association with temperature during the summer months.

2. Materials and methods 2.1. Variables As the dependent variable for our analyses, we used the natural logarithm of the monthly odds of live male and female infants in Sweden weighing less than 1500 g (i.e., the clinical definition of very low birth weight). We obtained the data from the Swedish Medical Birth Registry, aggregated by month to avoid problems with small numbers and attendant concerns about anonymity violations. We began our analyses with the first month of data availability (January, 1973). Our analyses ended with December, 2010, the most recent month for which we could obtain data at the time of our request. Swedish women delivered an average of 8240 live infants each month over the 456 test months. Of these, an average of 44 infants per month weighed less than 1500 g. Female infants had a risk (5.33 per 1000) similar to males (5.34 per 1000) of very low weight. Table 1 provides additional statistical detail concerning VLBW during the test period, broken out by the sex of the infant. Figs. 1 and 2 display the variation in the logarithm of the monthly odds of VLBW over the test period for males and females, respectively.

70

T. Hartig, R. Catalano / Health & Place 22 (2013) 68–74

Table 1 Descriptive statistics for the total number of singleton births and the number of very low birth weight ( o 1500 g) singleton births in Sweden for each month of the year, based on the 38 years 1973–2010, for male and female infants. Month

January February March April May June July August September October November December

Males born mean (SD)

Females born mean (SD)

Males o 1500 g mean (SD)

Females o1500 g mean (SD)

4238.34 4141.39 4755.61 4706.00 4607.61 4352.13 4434.32 4278.26 4153.76 3955.76 3602.11 3614.61

3980.61 3909.03 4489.55 4450.00 4346.97 4113.21 4171.47 4068.05 3931.76 3771.39 3398.53 3407.79

25.18 22.82 23.42 22.55 23.61 22.45 22.05 21.95 20.39 20.95 23.47 23.03

24.61 20.68 22.39 21.71 23.29 20.29 20.42 20.47 18.53 21.79 20.11 22.32

(481.44) (455.00) (528.86) (489.99) (465.68) (477.55) (491.96) (459.18) (434.73) (408.80) (403.45) (404.74)

(467.56) (416.40) (514.33) (490.97) (438.78) (458.41) (443.58) (429.77) (415.40) (399.59) (371.53) (392.13)

(6.79) (5.14) (6.15) (5.50) (5.27) (4.95) (6.63) (5.13) (5.13) (5.60) (6.06) (4.82)

(6.73) (5.29) (6.06) (7.03) (5.83) (5.60) (5.69) (4.85) (5.01) (5.97) (4.50) (5.72)

-4.00

-4.50

Logits

-5.00

-5.50

-6.00

-6.50 1

36

71

106 141 176 211 246 281 316 351 386 421 456

Months Fig. 1. Natural logarithm of the monthly odds of VLBW for male infants in Sweden during the 38 years 1973–2010 (N ¼ 456 months).

-4.00

-4.50

Table 2 Descriptive statistics for mean monthly temperature in Celsius in Sweden for each month of the year, based on the 38 years 1973–2010. Month

Minimum

Maximum

Mean

SD

January February March April May June July August September October November December

−11.22 −11.11 −4.52 2.08 7.84 11.58 14.68 12.97 8.71 3.30 −1.07 −8.05

2.59 3.72 4.16 7.35 12.70 16.98 19.88 19.49 14.62 9.89 5.96 4.11

−2.47 −2.67 0.25 4.70 10.29 14.44 16.84 15.91 11.47 6.64 2.12 −1.40

3.23 3.29 2.09 1.35 1.16 1.19 1.41 1.44 1.36 1.61 1.80 2.63

months during our test period. We included only locations for which at least 97% of the months had a valid value. Distributed from Mälmö in the south to Luleå in the north, these population centers lie outside of the relatively high, cool, and sparsely populated mountainous area that extends eastward from the border with Norway. The Swedish Meteorological and Hydrological Institute (SMHI) provided these data. Table 2 provides descriptive statistics for the observed temperature values. We derived the expected temperature values using times-series modeling (Box and Jenkins, 1970). As described in more detail below, the residuals of these models measure the degree to which an observed value differs from that expected from historical patterns (autocorrelation) in a series. Temperature covaries with other weather conditions. For example, in national climate data published by SMHI for the years 1860–2011, mean temperature for the summer season (June, July, and August) correlated inversely with mean precipitation; cooler summers had more rain (r ¼−0.27) (see http://www.smhi.se/ klimatdata/meteorologi/). Our focus on temperature here follows from the example set by Hartig et al. (2007) in their study of the association between summer temperature and the dispensation of anti-depressants to the Swedish population, and it simplifies the comparison of results across the studies. Our test equations also included a binary indicator variable for each of the three summer months. We scored the June variable, for example, 1 for each June and 0 otherwise. These three variables control the possible inverse association between summer months per se, and the likelihood of very low weight births.

Logits

-5.00

2.2. Design -5.50

-6.00

-6.50 1

36

71

106 141 176 211 246 281 316 351 386 421 456

Months Fig. 2. Natural logarithm of the monthly odds of VLBW for female infants in Sweden during the 38 years 1973–2010 (N ¼ 456 months).

As the focal independent variables for our analyses, we used the difference between expected and observed monthly mean temperatures (in Celsius) for June, July and August. The observed monthly means represent the mean of daily means. The daily means encompass the high and low temperatures for the day as well as readings taken at specified time points in the morning, afternoon and evening. The measurements came from 18 weather stations located near major population centers. Some locations did not have data for all

Our test turns on whether the observed likelihood of VLBW differs, as predicted by theory, from the values expected under the null hypothesis. Our null hypothesis holds that the odds of VLBW did not increase above expected levels when summer months had colder temperatures than expected. Such tests typically assume that the mean of the observed values represents the expected value of the dependent variable under the null hypothesis. The risk of VLBW, however, often exhibits autocorrelation, which can include trends, cycles (e.g. seasonality), and a tendency to remain elevated or depressed after high or low values. Autocorrelation complicates hypothesis tests because the mean of the observed values does not adequately represent the expected value of an autocorrelated series. Since Fisher's (1921) seminal work on crop yields, researchers have responded to this problem by expressing autocorrelation as an effect of earlier values, or lags, of the dependent variable itself. The residuals from an equation with the correctly identified lags exhibit no autocorrelation. The analyst can therefore add the independent variables to the equation to determine if their coefficients differ from zero in the hypothesized direction.

T. Hartig, R. Catalano / Health & Place 22 (2013) 68–74

Removing autocorrelation from the dependent variable before testing has the added benefit of reducing sources of a type I error. Removing autocorrelation precludes a spurious association resulting from third variables that exhibit the same trends, seasonality, or other patterns. 2.3. Analyses Our tests proceeded through the following steps.

71

August fall below their 95% confidence intervals (based on the reported standard errors). The coefficients for females have mixed signs (i.e., positive for July, negative otherwise) and none fall outside their 95% confidence intervals. Table 3 Coefficients and standard errors (SE) for predictors of the natural logarithm of the sex specific monthly odds of a very low weight birth in Sweden (N ¼ 456 months beginning January, 1973 and ending December, 2010). Males

1. We modeled the natural logarithm of the monthly odds of very low weight male or female births in Sweden as a function of a constant and the three summer month binary variables. 2. We inspected the residuals from Step 1 for autocorrelation. We implemented Fisher's strategy using methods developed by Box and Jenkins (1970) to identify and model autocorrelation. The strategy, Auto Regressive, Integrated, Moving Average (i.e., ARIMA) modeling, draws from a very large family of models available to empirically specify autocorrelation in time series. We chose ARIMA modeling because no approach has proved more accurate or efficient at detecting and modeling seasonality, a special case of higher order autocorrelation, than the Box–Jenkins methods (Fuller, 1996; Brockwell and Davis, 2002). 3. We applied the Box–Jenkins methods to monthly temperature to estimate expected values and residuals that gauge the degree to which each of the months differed from its expected value. 4. We specified the test equation by adding the residuals for June, July, and August estimated in Step 3 into the equation resulting from Step 2. 5. We estimated the equation resulting from Step 4 and inspected the error terms for autocorrelation. When we found any, we included additional ARIMA parameters and estimated the resulting equation. Our test equation appears as follows: 

vlbwt nbwt

e ¼ c þ ω1 X 1t þ ω2 X 2t þ ω3 X 3t þ ω4 X 4t þ ω5 X 5t þ ω6 X 6t þ

ð1−θBq Þ at ð1−ϕBp Þ

vlbwt represents the number of males or females weighing less than 1500 g born in month t. nbwt represents the number of males or females weighing more than 1499 g in month t. c represents a constant. X1t to X3t represent the three binary variables scored 1 for June, July, or August and 0 otherwise. X4t to X6t represent the June, July, and August residuals from the Box–Jenkins model for monthly mean temperature. We scored these variables 0 for months other than June, July, or August. ω0 through ω6 represent the estimated coefficients. θ represents the moving average parameter. ϕ represents the autoregressive parameter. B represents the “backshift operator” that yields the value of the series it conditions at time t-p for the autoregressive parameter or t-q for the moving average parameter. at represents the error term for month t. 3. Results Table 3 shows the results from the five steps above for male and female infants. The hypothesized inverse associations appear for males in all three summer months, but only those in June and

Constant June temperature residual June binary July temperature residual July binary August temperature residual August binary Autoregressive parameters

n

Females

Coefficient

SE

Coefficient

SE

−5.2101nn −0.0593nn −0.0436 −0.0289 −0.0848 −0.0453n −0.0071

0.0240 0.0282 0.0624 0.0259 0.0644 0.0271 0.0622

−5.2114nn −0.0357 −0.1116 0.0394 −0.1065 −0.0146 −0.0672

0.0258 0.0301 0.0724 0.0252 0.0735 0.0308 0.0723

B¼0.1945nn B12 ¼0.1652nn B36 ¼ 0.2007nn

0.0506 0.0509 0.0484

B¼ 0.1159nn B12 ¼ 0.2119nn B24 ¼ 0.2511nn

0.0495 0.0499 0.0488

p o 0.05; one-tailed test. po 0.01; one-tailed test.

nn

The augmented Dickey–Fuller test (Fuller,1996) detected no secular trends or sine waves in the odds of a very low weight birth among male or female infants. Our ARIMA models for these variables, therefore, did not include differencing. The best fitting models for males and females respectively appear as follows: Z mt ¼ c þ 1=ð1−∅BÞð1−∅B12 Þð1−∅B36 Þat Z f t ¼ c þ 1=ð1−∅BÞð1−∅B12 Þð1−∅B24 Þat The autoregressive parameters at t−1 suggest “short term” memory in which the odds of a very low weight birth at time t−1 predicts the odds at time t (see also Table 3). Both series also exhibited significant seasonality, as indicated by the autoregressive parameters at t−12 as well as at t−24 and t−36, for females and males respectively. Step 3 in our test yielded the following ARIMA model of our temperature variable: ∇12 Z t ¼

ð1−:3977BÞ ð1−:8427B12 Þ

at

Differencing at 12 months (i.e., month t subtracted from month t−12) as well as the autoregressive parameter at 12 (i.e., −0.8427) shows that, as expected, monthly average temperature exhibited strong seasonality. The moving average parameter (i.e., −0.3977) further suggests “short term” memory in which the differenced temperature value at time t−1 predicted the value at time t. The fact that the model includes neither differencing at 1 nor a constant shows that the temperature series exhibited no secular trends over the test period. We used the method described by Liu and Hudak (1992) and attributed to Tsay (1988) and others (Chang et al., 1988; Hillmer, 1984) to assess the likelihood that outliers inflated our confidence intervals, making our estimates inefficient and leading to false acceptance of the null for females and males in July. Controlling for outliers did not change our results. To provide perspective on the association we discovered for male infants, we rescored our June and August temperature variables to 1 for cold months (i.e., those with negatively signed values) and 0 otherwise and estimated our test equation with these rescored variables and without July temperature or the binary variables. Taking the antilog of the estimated coefficients suggests that a colder than expected June yielded a 13.6% increase

72

T. Hartig, R. Catalano / Health & Place 22 (2013) 68–74

in the odds of a very low weight male infant. A colder than expected August yielded a 5.4% increase. Estimating our test with colder than expected months rescored to 1 and warmer than expected months rescored to 0 had the additional benefit of assessing the sensitivity of our original results to our estimation, in Step 3, of the difference between observed and expected monthly temperature. The fact that the binary rescoring produced the same inference as our original test suggests that our results did not depend on our particular estimates of the difference between observed and expected temperature. Any model that separated months into roughly the same groupings of unexpectedly cold or warm as did our model would have led to the same inference.

4. Discussion The present results partially align with our constrained restoration hypothesis. Male infants born in a June or August with colder than expected temperatures had higher log odds of a very low weight birth. The same held for male infants born during July, though not to a statistically significant degree, perhaps because the extreme concentration of vacationing known to occur during that month released social resources that in other ways helped pregnant women deal with sources of stress. We did not predict male specific risk of VLBW in response to cold summers. We can, however, invoke, post hoc, natural selection in utero as a possible explanation (Catalano and Bruckner, 2006). Small males, particularly small male infants, die more frequently than other persons before and through reproductive life, despite greater maternal investment (Baird, 2009; Kristensen et al., 2007; Lummaa, 2001; Rickard et al., 2007). This disparity, moreover, increases in stressful environments (Catalano et al., 2005). Much literature therefore argues that natural selection has conserved mechanisms by which women spontaneously abort small male fetuses during stressful times (Baird, 2009; Forbes, 2005; Møller, 1997; Stearns, 1987; Trivers and Willard, 1973). The literature also argues that the prematurity associated with VLBW implies that many very low weight infants would have registered as spontaneous abortions before the advent of medical technology that saves premature, very small infants (Catalano et al., 2012). Should these arguments hold true, one would expect our finding that VLBW increases particularly among males during cold summers. The autoregressive parameters we discovered at 12, 24, and 36 months speak to seasonality in VLBW. We cannot say whether this means that cold weather per se constitutes a stressor for this population, or that other variables strongly correlated with ambient temperature, such as sunlight or workload, put women at greater risk. Swedes ordinarily have means readily at hand to maintain thermal comfort in the face of cold weather, such as wearing warmer clothing or simply staying indoors. We have therefore considered relatively cold summer temperature primarily as a condition that constrains restoration and not a stressor itself. It perhaps goes without saying that the 31.1 1C difference between the temperature extremes across the seasons in the period under study (−11.22 1C in the coldest January to 19.88 1C in the hottest July) far exceeds the roughly 5–7 1C difference seen within each of the summer months in our data set (see Table 2). Admittedly, some people may prefer cooler summer temperatures, and cooler temperatures might actually better serve some summer outdoor activities. We also acknowledge that if people cannot engage in restorative activities outdoors, they may look for alternative activities indoors that may also serve restoration. The question remains, however, whether those activities serve restoration as well as the preferred outdoor activities (Hug et al., 2009).

We assumed that relatively cold summer temperatures would cause a net reduction in restorative activities outdoors, and that this would echo in stress-related health outcomes such as VLBW. In sum, by referring to constrained restoration due to relatively cold summer weather, we provided a plausible a priori account for the associations between temperature deviations and the odds of VLBW uncovered in our analyses. A rival explanation must attribute the associations found to a phenomenon that co-varies with temperature deviations for the given months across the period within question and that at the same time does not exhibit autocorrelation. Arguably, one could take a more direct approach to studying the constraint of restoration due to weather variations. By repeatedly surveying late-term pregnant women with regard to their experiences of stress, restoration, and activities indoors and outdoors over the summer months, perhaps with a daily diary method (Kahneman et al., 2004), one could see whether reduced outdoor activities and constrained restoration with cooler temperatures predicted the subsequently recorded birth weight. We respect this point of view, but it does neglect some advantages with the approach taken in the present study. Not least, by looking at the experience of the population across time, we captured variation in the constraint of concern. When all in a cohort (e.g., of women in the final trimester during the summer) behave similarly in the face of the same constraint, a study may fail to adequately assess an association between the exposure and the target outcome (Rose, 1985). Our approach also had the advantage of prudence. Using immediately available data for the independent and dependent variables, we could at relatively low cost uncover initial evidence of an association between constrained restoration and VLBW. Researchers could understandably appreciate an initial estimate of effect before embarking on an expensive, individual-level study of a novel, hypothetical phenomenon such as the one investigated here. Aside from more directly assessing one or more of the linkages posited here among cool summer weather, reduced outdoor activity, and constrained restoration, further research could consider other outcomes potentially intermediate to very low birth weight, such as alcohol consumption or depression (cf. Field et al., 2006), that might occur as pregnant women try to cope with unrelieved chronic stress during cool summer months. Research can also investigate the effects of constrained restoration in the population more generally, as previously done with the dispensation of anti-depressants in Sweden (Hartig et al., 2007). Further research could also consider more complex formulations of the temperature variable. In the present study we used a simple formulation of the temperature variable for the given month, looking only at the possibility of an effect on birth weights in the same month and not at lagged effects. More complex formulations could address the issue of cumulative effects. How people experience the constraint of restoration in a given summer month may reflect on what they experienced in the preceding months. Cool summer months may have more potent effects if they follow spring months that also had below normal temperatures. Another question for further research concerns the degree to which other meteorological variables may harm health by constraining restoration. As mentioned earlier, Eliasson et al. (2007) found that temperature had a stronger association with the number of people present in different urban outdoor spaces than did either the amount of cloud cover or wind speed. Those additional variables did nonetheless contribute to the explanation of presence outdoors, and one could ask whether they thus could harm health via a behavioral pathway, alone or in interaction with temperature.

T. Hartig, R. Catalano / Health & Place 22 (2013) 68–74

Finally, further research could also consider meteorological conditions that interfere with outdoor activities in other geographical locations. For example, in some regions unusually high summer temperatures may keep people indoors and less active, against their will (cf. Townsend et al., 2003; Lafortezza et al., 2009; Hipp and Ogunseitan, 2011). This issue, much like relatively cool summer conditions in some regions, may become increasingly salient given global warming. In conclusion, our theoretical arguments and empirical results illustrate how weather may figure in stress-related outcomes not only as a stressor, but also as a constraint on restoration. Moreover, our results agree in broad outline with those of a previous study, in which the dispensation of antidepressants in Sweden varied inversely with mean monthly temperature for July looking across the years 1991 through 1998 (Hartig et al., 2007). Together, these two studies well exemplify the potential utility of the constrained restoration concept for guiding studies on relations among stress, restoration, environmental circumstances, and health.

Acknowledgments We thank Ingela Bengtsson and Ingalill Halvarsson for helping us to locate statistics and governmental documents concerning Swedish vacation legislation and practices. We also thank the two anonymous reviewers for their helpful comments. References American College of Obstetricians and Gynecologists, 2001. Air travel during pregnancy. ACOG Committee Opinion no. 264. Obstetrics and Gynecology 98, 1187–1188. Andra Lagutskottet, 1951. Utlåtande i anledning av dels Kungl. Maj:ts proposition med förslag till lag angående ändring i lagen den 29 Juni 1945 (nr. 420) om semester, m.m., dels ock i ämnet väckta motioner (report in connection in part with the Government's proposition, with proposals for the law concerning change in the vacation law of 29 June 1945 (nr. 420), and more, and in part with motions raised in the matter) (Utlåtande nr. 28). Sveriges Riksdag, Stockholm. Andra Lagutskottet, 1953. Utlåtande i anledning av väckta motioner om viss ändring i 12 § lagen om semester (Report in connection with motions raised concerning change in the 12 § of the vacation law) (Utlåtande nr. 20). Sveriges Riksdag, Stockholm. Baird, D.D., 2009. The gestational timing of pregnancy loss: adaptive strategy? American Journal of Human Biology 21, 725–727. Box, G., Jenkins, G., 1970. Time Series Analysis: Forecasting and Control. HoldenDay, San Francisco. Brockwell, P., Davis, R., 2002. Introduction to Time Series and Forecasting, 2nd ed. Springer, New York. Catalano, R., Bruckner, T., 2006. Secondary sex ratios and male lifespan: damaged or culled cohorts. Proceedings of the National Academy of Sciences of the United States of America 103, 1639–1643. Catalano, R., Bruckner, T., Anderson, B., Gould, J., 2005. Fetal death sex ratios: A test of the economic stress hypothesis. International Journal of Epidemiology 34, 944–948. Catalano, R., Goodman, J., Margerison-Zilko, C., Saxton, K., Anderson, E., Epstein, M., 2012. Selection against small males in utero: A test of the Wells hypothesis. Human Reproduction 27, 1202–1208. Chang, I., Tiao, G., Chen, C., 1988. Estimation of time series parameters in the presence of outliers. Technometrics 30, 193–204. Cohen, S., Kessler, R.C., Gordon, L.U., 1997. Strategies for measuring stress in studies of psychiatric and physical disorders. In: Cohen, S., Kessler, R.C., Gordon, L.U. (Eds.), Measuring Stress: A Guide for Health and Social Scientists. Oxford University Press, Oxford, pp. 3–26. Crouse, D., Cassady, G., 1994. The small-for-gestational-age infant. In: Avery, G., Fletcher, M., MacDonald, M. (Eds.), Neonatology: Pathophysiology and Management of the Newborn, 4th ed. Lippincott, Philadelphia, PA, pp. 369–387. Coussons-Read, M.E., Okun, M.L., Nettles, C.D., 2007. Psychosocial stress increases inflammatory markers and alters cytokine production across pregnancy. Brain, Behavior and Immunity 21, 343–350. Da Costa, D., Rippen, N., Dritsa, M., Ring, A., 2003. Self-reported leisure-time physical activity during pregnancy and relationship to psychological wellbeing. Journal of Psychosomatic Obstetrics and Gynecology 24, 111–119. Donovan, G.H., Michael, Y.L., Butry, D.T., Sullivan, A.D., Chase, J.M., 2011. Urban trees and the risk of poor birth outcomes. Health & Place 17, 390–393. Eliasson, I., Knez, I., Westerberg, U., Thorsson, S., Lindberg, F., 2007. Climate and behaviour in a Nordic city. Landscape and Urban Planning 82, 72–84.

73

Ericson, B., Gustaffson, S., 1977. Nya Semesterlagen (The New Vacation Law). Tidens förlag, Stockholm. Field, T., Diego, M., Hernandez-Reif, M., 2006. Prenatal depression effects on the fetus and newborn: a review. Infant Behavior Development 29, 445–455. Fisher, R.A., 1921. Studies in crop variation. I. An examination of the yield of dressed grain from Broadbalk. Journal of Agricultural Science 11 (02), 107–135. Forbes, S., 2005. A Natural History of Families. Princeton University Press, Princeton, NJ. Fuller, W., 1996. Introduction to Statistical Time Series, 2nd ed. Wiley, Hoboken, NJ. Gäddlin, P-O., 2011. Follow-up studies of very low birthweight children in Sweden. Acta Paediatrica 100, 940–949. Hartig, T., 2007. Three steps to understanding restorative environments as health resources. In: Ward Thompson, C., Travlou, P. (Eds.), Open Space: People Space. Taylor & Francis, London, pp. 163–179. Hartig, T., Catalano, R., Ong, M., 2007. Cold summer weather, constrained restoration, and the use of antidepressants in Sweden. Journal of Environmental Psychology 27, 107–116. Hillmer, S., 1984. Monitoring and adjusting forecasts in the presence of additive outliers. Journal of Forecasting 3, 205–215. Hipp, A.J., Ogunseitan, O.A., 2011. Effect of environmental conditions on perceived psychological restorativeness of coastal parks. Journal of Environmental Psychology 31, 421–429. Hobel, C., Culhane, J., 2003. Role of psychosocial and nutritional stress on poor pregnancy outcome. Journal of Nutrition 133, 1709S–1717S. Hobel, C.J., Goldstein, A., Barrett, E.S., 2008. Psychosocial stress and pregnancy outcome. Clinical Obstetrics and Gynecology 51, 333–348. Hogue, C.J.R., Hoffman, S., Hatch, M.C., 2001. Stress and preterm delivery: a conceptual framework. Pediatric and Perinatal Epidemiology 15 (Supplement), 30–40. Hug, S.M., Hartig, T., Hansmann, R., Seeland, K., Hornung, R., 2009. Restorative qualities of indoor and outdoor exercise environments as predictors of exercise frequency. Health & Place 15, 971–980. Kahneman, D., Krueger, A.B., Schkade, D., Schwarz, N., Stone, A.A., 2004. A survey method for characterizing daily life experience: The Day Reconstruction Method (DRM). Science 306, 1776–1780. Katz, V.L., 2012. Work and work-related stress in pregnancy. Clinical Obstetrics and Gynecology 55, 765–773. Kunglig Majestät, 1951. Kungl. Maj:ts proposition till Riksdagen med förslag till lag angående ändring i lagen den 29 Juni 1945 (nr. 420) om semester, m.m. (The Government's proposition with proposals for the law concerning change in the vacation law of 29 June 1945 (nr. 420), and more) (Proposition nr. 72). Sveriges Riksdag, Stockholm. Kristensen, S., Salihu, H.M., Keith, L.G., Kirby, R.S., Fowler, K.B., Pass, M.A.B., 2007. SGA subtypes and mortality risk among singleton births. Early Human Development 83, 99–105. Lafortezza, R., Carrus, G., Sanesi, G., Davies, C., 2009. Benefits and well-being perceived by people visiting green spaces in periods of heat stress. Urban Forestry and Urban Greening 8, 97–108. Liu, L.M., Hudak, G.B., 1992. Forecasting and Time Series Analysis Using the SCA Statistical System. Scientific Computing Associates, Oak Brook, IL. Lovallo, W.R., 2005. Stress and Health: Biological and Psychological Interactions. SAGE, London. Lummaa, V., 2001. Reproductive investment in pre-industrial humans: the consequences of offspring number, gender and survival. Proceedings of the Royal Society B Biology 268, 1977–1983. Matthews, C.E., Freedson, P.S., Hebert, J.R., Stanek III, E.J., Merriam, P.A., Rosal, M.C., Ebbeling, C.B., Ockene, I.S., 2001. Seasonal variation in household, occupational, and leisure time physical activity: longitudinal analyses from the Seasonal Variation of Blood Cholesterol Study. American Journal of Epidemiology 153, 172–183. McMichael, A.J., Woodruff, R.E., Hales, S., 2006. Climate change and human health: present and future risks. Lancet 367, 859–869. Møller, A.P., 1997. Developmental selection against developmentally unstable offspring and sexual selection. Journal of Theoretical Biology 185, 415–422. Otchet, F., Carey, M.S., Adam, L., 1999. General health and psychological symptom status in pregnancy and the puerperium: what is normal? Obstetrics and Gynecology 94, 935–941. Pivarnik, J.M., Reeves, M.J., Rafferty, A.P., 2003. Seasonal variation in adult leisuretime physical activity. Medicine and Science in Sports and Exercise 35, 1004–1008. Rickard, I., Russell, A., Lummaa, V., 2007. Producing sons reduces lifetime reproductive success of subsequent offspring in pre-industrial Finns. Proceedings of the Royal Society B Biology 274, 2981–2988. Rook, K., Dooley, D., Catalano, R., 1991. Stress transmission: The effects of husband's job stressors on the emotional health of their wives. Journal of Marriage and the Family 53, 165–177. Rose, G., 1985. Sick individuals and sick populations. International Journal of Epidemiology 14, 32–38. Statens Offentliga Utredningar, 1944. Betänkande med Förslag till Ändrad Semesterlagstiftning (Report with Proposals for Changing Vacation Legislation) (SOU 1944:59). Socialdepartementet, Stockholm. Statens Offentliga Utredningar, 1964. Friluftslivet i Sverige, Del 1: Utgångsläge och Utvecklingstendenser 1962 Års Fritidsutredning (Outdoor Recreation in Sweden, Part 1: Current Conditions and Development Tendencies 1962 Leisure Inquiry) (SOU 1964:47). Kommunikationsdepartementet, Stockholm.

74

T. Hartig, R. Catalano / Health & Place 22 (2013) 68–74

Statens Offentliga Utredningar, 1967. Semesterspridning (Vacation Distribution) (SOU 1967:61). Socialdepartementet, Stockholm. Statens Offentliga Utredningar, 1975. Fem Veckors Semester (Five Weeks Vacation) (SOU 1975:88). Arbetsmarknadsdepartementet, Stockholm. Statistiska Centralbyrån, 2004. Fritid 1976–2002 (Leisure Activities 1976-2002) (Levnadsförhållanden Rapport nr. 103). Statistiska Centralbyrån, Stockholm. Stearns, S.C., 1987. The selection-arena hypothesis. In: Stearns, S.C. (Ed.), The Evolution of Sex and Its Consequences. Birkhauser Verlag, Basel, pp. 299–311. Thorsson, S., Lindqvist, M., Lindqvist, S., 2004. Thermal bioclimatic conditions and patterns of behaviour in an urban park. International Journal of Biometeorology 48, 149–156. Townsend, M., Mahoney, M., Jones, J.-A., Ball, K., Salmon, J., Finch, C.F., 2003. Too hot to trot? Exploring potential links between climate change, physical activity, and health. Journal of Science and Medicine in Sport 6, 260–265.

Trivers, R.L., Willard, D.E., 1973. Natural selection of parental ability to vary the sex ratio of offspring. Science 179, 90–92. Tsay, R., 1988. Outliers, level shifts, and variance changes in time series. Journal of Forecasting 7, 1–20. Wadhwa, P.D., Glynn, L., Hobel, C.J., Garite, T.J., Porto, M., Chicz-DeMet, A., Wiglesworth, A.K., Sandman, C.A., 2002. Behavioral perinatology: biobehavioral processes in human fetal development. Regulatory Peptides 108, 149–157. Wadhwa, P.D., Culhane, J.F., Rauh, V., Barve, S.S., 2001. Stress and preterm birth: neuroendocrine, immune/inflammatory, and vascular mechanisms. Maternal and Child Health Journal 5, 119–125.