Induced resistance of host tree foliage during and after a natural insect outbreak


Pekka Kaitaniemi, Section of Ecology, Department of Biology, University of Turku FIN-20014 Turku, Finland. Tel. +358 2333 5072. Fax: +358 2333 6550. E-mail:


1. Plant resistance against insect herbivores often increases after experimental damage to foliage, but few studies have obtained field estimates of the effect of induced resistance on insect populations during and after a natural insect outbreak.

2. This study measured the effect of quality of the host tree, mountain birch (Betula pubescens ssp. czerepanovii), on the periodically fluctuating folivore Epirrita autumnata (Lepidoptera, Geometridae) during peak and postpeak years of an outbreak in Finnish Lapland. Comparisons were made both within and between study sites to assess host plant quality, and thereby the effect of delayed induced resistance (DIR).

3. In within-site comparisons, a set of experimental trees was defoliated by wild larvae in the peak year of the outbreak, whereas control trees were protected from defoliation by spraying with an insecticide. The effect of host plant quality was quantified in the following year by measuring the pupal mass of E. autumnata larvae reared in enclosures on these trees.

4. In between-site comparisons, the sizes of pheromone-trapped males were measured at both outbreak and low density sites during the progress of the outbreak. The size of trapped males was subsequently used to estimate the corresponding fecundity of females at the same sites.

5. Pupal mass of E. autumnata reared on trees defoliated in the previous year was 0–10% lower than on those trees protected from defoliation by the insecticide. Field-collected adults indicated a similar pattern: they were smaller at outbreak sites than at low-density sites, and the size reached its minimum in the post-peak year. However, the estimated loss of reproductive capacity of females resulting from DIR was too small to be the sole explanation for the termination of the outbreak.

6. Whether the weak DIR response in this system was a characteristic of the 1990s outbreak alone remains unclear, because different terminating agents may be important for different individual outbreak peaks. During this outbreak, larval parasitism and developmental asynchrony between larvae and birch were probably more important reasons for population collapse than DIR.


The relative role of different factors behind cyclicity in population density of forest Lepidoptera has remained controversial (Myers 1988; Berryman 1996). One of the possible driving factors behind population cycles is herbivore-induced, density-dependent deterioration in plant quality. If such induced resistance has delayed detrimental effects on survival or fecundity of insects in years following large scale defoliation, it can suppress the population for some years, until its efficiency is mitigated and population density can start to rise again (Haukioja 1980, 1982). The reduction in fecundity of Lepidopteran species has been observed during outbreaks in the field (Mason, Beckwith & Paul 1977; Carter, Ravlin & McManus 1991), and such changes have been attributed to changes in foliage quality (Baltensweiler & Fischlin 1988), but, in general, data concerning the effect of induced changes in host quality on natural insect populations are lacking (Karban & Baldwin 1997).

Increased plant resistance, interpreted from decreased performance of herbivores, has often been demonstrated after experimental damage to foliage (for reviews see Haukioja & Neuvonen 1987; Karban & Myers 1989; Haukioja 1991; Tallamy & Raupp 1991); for example, in numerous experiments with birches (Betula spp.), defoliation has been shown to trigger both delayed and rapid forms of induced resistance [delayed induced resistance (DIR) and rapid induced resistance (RIR), respectively] in foliage (Neuvonen & Haukioja 1991). Because RIR affects the current generation of herbivores, it tends to stabilize insect population dynamics, whereas DIR, because of its delayed density dependence on successive generations, can contribute to cyclic fluctuations in insect density (Haukioja 1982).

However, the possible role of insect–host plant interactions in insect population dynamics is more complex than RIR or DIR alone; for example, insects may be able to avoid resistant plants in the field (e.g. Sork, Stowe & Hochwender 1993), insect performance may be affected by the effect of defoliation on synchrony between leaf flush and eclosion of larvae (Kaitaniemi, Ruohomäki & Haukioja 1997b), and insects may have feeding habits which help them to overcome plant resistance partially or completely (Tallamy 1985; Dussourd & Eisner 1987; Roland & Myers 1987; Haukioja et al. 1988b; Sagers 1992; Kaitaniemi, Ruohomäki & Haukioja 1997a). Kaitaniemi et al. (1997a) have recently shown that the larvae of Epirrita autumnata (Bkh.) (Lepidoptera, Geometridae) were able to alleviate the DIR response of birch by consuming, simultaneously with defoliation, the primordial apical buds developing for the next year's growth. Accordingly, experimental studies must be accompanied by field observations to estimate the total effect of different factors on insect populations. Understanding the contribution of different factors is increasingly important as the climate change may render wider areas susceptible to outbreaks of E. autumnata (Ruohomäki et al. 1997; Virtanen, Neuvonen & Nikula 1998) and other species.

The present study used two complementary methods to evaluate the importance of DIR and other food plant-related traits on the performance of E. autumnata during a natural population peak, during the onset of population decline, and during a low-density phase in Finnish Lapland between 1992 and 1997. First, in 1993–97 the sizes were recorded of pheromone-trapped E. autumnata males caught from study plots located both at outbreak and low-density sites. Second, in 1994, the year following the first peak year, foliage quality was bioassayed directly at outbreak sites by rearing larvae on trees which had either been defoliated by wild larvae or been protected from defoliation by insecticide spraying in the previous year.

Materials and methods

Study organisms and sites

Epirrita autumnata is a spring folivore which overwinters in the egg stage. The larvae hatch at the time of bud burst of birch, and complete their growth within about 6 weeks. Adults emerge in August or September. Populations of E. autumnata fluctuate periodically in northern Fennoscandia, reaching outbreak densities at 9- to 10-year intervals and sweeping across mature birch forests (Tenow 1972; Haukioja et al. 1988a; Bylund 1995; Ruohomäki et al. 1997). Because of the extensive defoliation it causes during such outbreaks, E. autumnata is the most destructive herbivore of its staple food plant, mountain birch [Betula pubescens ssp. czerepanovii (Orlova) Hämet-ahti, syn. B. pubescens ssp. tortuosa (Ledeb.) Nyman; Tenow 1972; Kallio & Lehtonen 1973).

The performance of E. autumnata is largely determined by the quality of the foliage it consumes, and is reflected in the final larval mass or pupal mass (Neuvonen & Haukioja 1991; Tammaru 1998). Pupal mass is often used as the measure of individual fitness, because it is the main determinant of potential fecundity in females (Haukioja & Neuvonen 1985a; Tammaru, Kaitaniemi & Ruohomäki 1996). The strong correlation between pupal mass and fecundity apparently persists also in the field conditions as a result of restricted importance of adult feeding and short adult lifespan in this species (Tammaru et al. 1996).

This study was conducted in the central part of Finnish Lapland at 10 outbreak sites and 12 low density sites of about 1 ha each (Fig. 1). The sites were characterized by mixed Norway spruce [Picea abies (L.) Karsten], Scots pine (Pinus sylvestris L.) and birch forest. The area belongs to the interzone of two hybridizing subspecies of white birch, Betula pubescens ssp. czerepanovii and pubescens (Ehrh.) (Kallio, Niemi & Sulkinoja 1983). The ground layer of the sites was typically covered by the dwarf shrubs Vaccinium myrtillus (L.), V. vitis-idaea (L.), V. uliginosum (L.) and Empetrum nigrum (L.). Other deciduous woody plants, dwarf birch [B. nana (L.)], and Salix spp., were present occasionally.

Figure 1.

Location of study sites in the outbreak area in northern Finland.

Estimation of larval density and damage to foliage

A density index (see Ruohomäki & Haukioja 1992) was used as a quick method to describe larval density at study sites. The density index was calculated as the number of larvae found per 10-min search at each site. At least 10 haphazardly selected birch trees, both short trees and the lowest branches of tall trees, were inspected for larvae for up to 30 s by the same person at all sites and in all years. A majority of larvae were at the third instar stage when densities were estimated.

The applicability of this density index in estimating absolute larval densities was assessed at 15 sites, where absolute larval densities (number of larvae per amount of foliage in birches) were also determined. The results of these density measurements were plotted against the density index, and a significant correlation was found (r = 0·61, P = 0·02). However, instead of using the density index as a continuous variable, it was used to divide the study sites into outbreak and low density classes, because at very high densities the number of simultaneously found larvae often exceeded the number that could be instantly counted. A minimum of 90 larvae/10 min was considered as sufficient for a site to have outbreak density, whereas sites having densities less than 50 larvae/10 min had a distinguishably lower density and degree of defoliation.

During the peak years 1993 and 1994, the degree of defoliation was visually estimated from 15 trees at six outbreak sites. To obtain defoliation estimates for 1992, feeding marks in leaf litter were used; samples of leaf litter were collected haphazardly from an area of a few thousand square meters at six outbreak sites in the springs of 1993–95 (10 or more plots mixed in one sample each year). Thus, for 1993 and 1994 figures were available for the degree of defoliation both directly estimated from the trees and taken from the leaf litter samples in the next spring (50 leaves/ sample). These data were used to construct a regression equation to estimate site-specific defoliation degrees for 1992 from the leaf litter samples collected in spring 1993.

Within-site experiment: effect of foliage quality on pupal mass of e. autumnata

The direct effect of defoliation-induced changes in foliage quality on E. autumnata larvae was measured by rearing larvae on trees at four outbreak sites in 1994 (sites 1, 4, 5 and 10; Fig. 1), the year following the first peak year. At each site, 30 mature 1·5–3·0 m high trees, located 5–50 m apart, were selected on 29 May in the previous year 1993. At that time E. autumnata larvae were hatching but had not yet caused significant damage to flushing leaves. At each site, 15 randomly assigned trees out of the 30 trees were sprayed with insecticide to prevent current-year defoliation. The insecticide used was a pyrethroid (Folition, Bayer, Germany), containing fenitrothion as the killing agent (and xylene as a solvent). The spraying concentration used was 5 mL in 10 L water. The other 15 trees were sprayed with water only. Defoliation of the insecticide-sprayed trees remained below 5% in the year when the spraying was conducted, and was caused by late-season herbivores. Hence the trees fell into two treatment groups: natural defoliation vs. insecticide-sprayed control trees.

The performance of E. autumnata larvae was measured in the following year 1994, by rearing 20 larvae (belonging to 20 different broods) from egg to pupa in a mesh bag in each tree (for more methodological details, see Haukioja & Neuvonen 1985b). On 21 May, a single 60-cm-long top branch of each tree was enclosed in a mesh bag with a similar mixture of E. autumnata eggs. Parents of the test larvae had been collected as third or fourth instar larvae from both outbreak and low density sites within the area of this study. An equal proportion of eggs from both origins were used for rearings at each site. The birch leaves enclosed in the mesh bags were in excess of consumption to prevent starvation.

At the end of the larval period, the mesh bags were checked daily for larvae approaching pupation; these were removed and allowed to pupate in 48-mL plastic vials (1–2 larvae per vial) half-filled with moist Sphagnum moss. Pupal mass, measured a fortnight after pupation, was used to assess the effect of foliage quality on E. autumnata performance. A separate study showed that the previous year insecticide spraying per se had no significant effect on the pupal mass (or survival) of larvae on sprayed trees in 1994 (Kaitaniemi et al. 1997a).

Between-site experiment: adult size of e. autumnata during and after the outbreak

Sizes of adult E. autumnata males were used to estimate the effect of foliage quality on wild larvae. Pheromone traps with synthesized E. autumnata pheromone (Zhu et al. 1995) were used to catch male moths. The trappings were carried out in the autumns of 1993–97, scheduling the trapping period each year to match the estimated peak of the flight period. For practical reasons, a slightly different combination of sites (Fig. 1) was used in different years. The outbreak sites 1, 4, 5, 6 and 10 were sampled in all years and the additional outbreak sites were: 2 and 8 in 1993; 2 in 1994; and 2, 7, 8 and 9 in all other years. The low density sites were: 11, 15–19, 21 and 22 in 1993; 11 and 22 in 1994; 11–15, 18–20 and 22 in both 1995 and 1996; 12–14 and 18 in 1997.

To assess the magnitude of normal year-to-year variation in male size, which could confound the potential effects of host plant quality, pheromone trappings were conducted in 1995–97 also at eight sites in an area of 150 km2 in south-western Finland near Turku. This area has no E. autumnata outbreaks, and is located several hundred kilometres away from the current outbreak area.

The wings and body were often damaged in the pheromone traps, so the hind-leg femur length of the males was used as a measure of size. The femur lengths of samples of 30 males from each site were measured by microscope to the nearest 0·03 mm.

Conversion of male size to female fecundity

The pupal masses of E. autumnata males and females are similarly affected by environmental conditions (Fig. 2, see also Tammaru 1998). It was thus possible to use the size of males trapped at the study sites to estimate the corresponding size and maximum fecundity of females at the same sites, and to evaluate the possible reduction in fecundity of females at the outbreak sites. To do this, male femur length (in mm) was first transformed into male pupal mass (in mg) by a regression equation based on laboratory-reared animals: male pupal mass = 46·0 × (male femur length) − 41·1 (R2 = 0·71, N = 27). The resulting male pupal mass was transformed into female pupal mass by the equation given in Fig. 2, and female pupal mass was transformed into egg number by a regression equation taken from Tammaru et al. (1996): egg number of a female = 2·93 × (female pupal mass) − 101·9 (R2 = 0·77, N = 26).

Figure 2.

Relationship between pupal mass of male and female E. autumnata reared on same foliage in identical conditions. Each dot represents average pupal mass of 5–15 males and 5–15 females.

To elucidate the typical low-density phase fecundity of E. autumnata, the average egg number of E. autumnata females originating from five low density phase sites, some 250 km north of the outbreak area (Table 1), was measured. These sites have not experienced outbreaks during the last 30 years. All five data sets, including three different years, yielded fairly similar egg numbers, and averaged 129 eggs with 95% confidence limits being 119–139 eggs.

Table 1.  Average egg numbers of field-grown Epirrita autumnata females originating from low-density-phase populations in Finnish Lapland
OriginNEgg number
(95% CL)
Wild adult females collected from
a field site in 1993
13138 (115–161)
Copulating females collected from
a field site in 1993
17135 (124–146)
Wild larvae collected from a field
site 2 days before pupation
in 1992, fed with host tree leaves
18119 (100–138)
Wild larvae collected from a field
site 2 days before pupation in 1992,
fed with host tree leaves
22128 (110–146)
Singly grown larvae from mesh bag
rearings in 1991
13126 (106–146)

Statistical analyses

The data were analysed using the SAS statistical package. Because of missing cells in the data, Proc Mixed was used to test the effect of larval density on size of pheromone-trapped males (Littell et al. 1996). In the analyses, year and density were considered as fixed effects, and the tests were based on the first-order autoregressive covariance structure of data (Littell et al. 1996).

In the within-site experiment, the effect of treatment on pupal mass of E. autumnata was analysed using an analysis of variance (anova). One water-sprayed and two insecticide-sprayed trees at site 10 were excluded because they had been damaged by a by-passing logging machine. As a result of the low number of sites in the experiment, the nested anova was used as suggested by McKone & Lively (1993) for experiments conducted at multiple, but few randomly chosen sites (i.e. sites that represent a sample from a larger population of sites). This method allowed the effect of the treatment itself to be better detected. In this method, the effect of treatment was tested separately for each site by nesting the treatment within each site. Normality of the data and homoscedasticity of the variances were confirmed with Shapiro–Wilk's tests and Cochran's tests, respectively.


The course of the outbreak

High larval densities were observed in the study area for the first time in the summer of 1992 (reported by local newspapers); the estimates of defoliation degree, based on feeding marks in leaf litter, likewise suggested a relatively high degree of defoliation already in 1992 (Fig. 3). The highest larval densities were observed in 1993 when the average of larval density index was 156 larvae found/10 min at outbreak sites, and 25 larvae/10 min at low density sites. In 1994 the values were 80 and 32 larvae/10 min, respectively. In 1995 densities had clearly declined at both outbreak (18 larvae/10 min) and at low-density sites (5 larvae/ 10 min), and remained below 4 larvae/10 min in both 1996 and 1997.

Figure 3.

Correspondence between defoliation estimates based on leaf litter and those obtained directly from trees. Numbers indicate study sites according to Fig. 1.

Effects of previous-year defoliation on pupal mass of e. autumnata

At three out of four outbreak sites (4, 5 and 10), the pupal mass of E. autumnata was lower in previously defoliated trees than in insecticide-sprayed trees, although significantly so only at site 4 (Table 2). The true differences may be slightly underestimated, because defoliation occurring in 1992 could have affected foliage quality of insecticide-sprayed trees. However, observations suggest that the DIR response loses most of its efficacy by the second summer after defoliation (Kaitaniemi, Neuvonen & Nyyssönen 1999), implying that, if DIR was present, it should have been weak in the insecticide-sprayed trees. Therefore, trees with a ‘strong’ DIR may have been compared with trees with a ‘weak’ DIR.

Table 2. anova table for differences in the pupal mass of Epirrita autumnata reared on birch which were either defoliated by larvae or protected from defoliation by an insecticide in the previous year. Hierarchical anova for three sites (4, 5, 10) according to McKone & Lively (1993)
 Pupal mass ± SEanova
Source (site)InsecticideDefoliationd.f.MSFP
  • Note that these values have been corrected for the effect of sex, and represent intermediate values compared to true male or female pupal mass.

Def(4)70·5 ± 2·869·2 ± 2·5112·470·100·751
Def(10)74·5 ± 3·668·8 ± 3·01213·071·730·192
Def(5)73·3 ± 3·165·0 ± 2·51492·853·990·049
Site  245·750·370·691
Error  80123·41  

At the fourth site (1) pupal mass was 65 mg in both defoliated and sprayed trees (reported in Kaitaniemi et al. 1997a), possibly as a result of the ameliorating effect on foliage quality of bud consumption by larvae at this site (see Kaitaniemi et al. 1997a). Bud consumption was not observed in the other outbreak sites.

Adult size of wild e. autumnata

Measurements of the femur length of pheromone-trapped males gave a similar picture of the effect of larval growth conditions: males were significantly smaller (F1,15 = 7·16, P = 0·02) at outbreak than at low-density sites during all the study years (Table 3). The smallest size was observed in 1994, the year after peak density, and it corresponded to females having 72 mg pupal mass or 108 eggs. The differences between years were also significant (F4,41 = 9·06, P = 0·0001): the largest size was achieved at low density sites 2 years after the outbreak in 1996, corresponding to 78 mg female pupal mass or 128 eggs. This maximum value is very close to values observed during the low density phase before the outbreak years (Table 1).

Table 3.  Durations of pheromone trappings, average femur lengths of Epirrita autumnata males, 95% confidence limits (CL), and average numbers of males measured (N) at each site: (a) in outbreak and low density sites within an outbreak area in northern Finland (1993, 1994 were outbreak years); and (b) in low density area in south-western Finland (outside outbreak area)
 OutbreakLow density
YearTrapping periodFemur (mm)95% CLNFemur (mm)95% CLN
  • The actual periods may have been shorter because of saturation of traps by the moth.

(a) Outbreak area
199322 Aug–08 Sep 2·402·37–2·43302·452·41–2·4830
199431 Aug–09 Sep 2·382·34–2·41302·442·39–2·4930
199523 Aug–04 Sep 2·462·43–2·48242·472·44–2·4924
199625 Aug–10 Sep 2·492·45–2·51242·512·49–2·5422
199723 Aug–08 Sep 2·482·45–2·51232·502·46–2·5431
(b) Low density area
199514 Sep–28 Sep 2·592·58–2·6127   
199619 Sep–07 Oct 2·592·57–2·6130   
199718 Sep–10 Oct 2·592·55–2·6228   

However, DIR was not strong enough to cause a significant ‘density × year’ interaction (F4,41 = 0·94, P = 0·45), which would have resulted from a particularly large size reduction after the peak year. The effect of DIR may have been partly confounded by a natural size difference between outbreak and low density sites: during 1996 and 1997 there still was a 0·02 mm difference, although otherwise the year-to-year figures seemed to be settling. An alternative explanation could be that DIR was still active, but this may be unlikely because the effect of DIR has been found largely to disappear within 2 years after even severe manual defoliation (Kaitaniemi et al. 1999).

In the area with no outbreaks in south-western Finland there was no detectable year-to-year size variation, and the average femur length was 2·59 mm in all 3 years (Table 3), suggesting that the average size remains largely constant during the low density phase.


This study was the first to assess the role of foliage quality of B. pubescens during a natural E. autumnata outbreak, and one of few field studies assessing food quality during an outbreak of any forest lepidopteran. Previous studies in our system had suggested that DIR can have a strong effect on the performance of E. autumnata in the year following experimental defoliation, although the efficacy of DIR has, in general, varied considerably among individual years and experiments (Ruohomäki et al. 1992). Experimentally induced larval defoliation, in particular, has seemed to induce a strong DIR response, and generally defoliation degrees approaching 50% of the canopy have been found to be sufficient (Haukioja & Neuvonen 1985b).

Both the approaches adopted by the present study, rearing larvae on trees in the field and monitoring the size of adults in the field, gave similar results regarding the suitability of foliage and other growth conditions for E. autumnata at outbreak sites. They suggested that quality of foliage decreased after defoliation—a result that was also reflected in the biochemical traits of leaves (Kaitaniemi et al. 1998). However, the reduction of adult size at outbreak sites was minor compared to size at low density sites, trees or years. During and after the outbreak, the estimated fecundity of females was maximally only 23 eggs (18%) below the average 129 eggs observed in the low density phase. The observed size reduction also had no consequences for the ability of males to fertilize the eggs of females (Haukioja & Neuvonen 1985a).

Consequently, the observed reduction in fecundity resulting from previous defoliation was presumably not the reason why the present peak in density terminated in 1994. This suggests other causes for the low survival rate of E. autumnata in 1994. That effect may have been a general one, because populations at both high and low density sites decreased simultaneously. The effects of parasitism will be treated separately in the future (K. Ruohomäki & P. Kaitaniemi, unpublished information), but it is worth mentioning here that the parasitism rates of larvae in 1993 (Ruohomäki et al. 1996) and in 1994 were not very high (below 50% at all sites; this does not include parasitism that occurred after the third larval instar). Neither were any signs of disease observed, which is in accordance with Bylund's (1995) observations, suggesting that diseases seldom infect E. autumnata during outbreaks. As discussed below, this leaves open two possible reasons related to food quality: the poor foliage quality in 1994 as a result of to the cold summer of 1993, and the poor synchrony of larval hatch and birch budbreak in the spring of 1994.

In the spring after a cold summer, the mountain birch leaves remain smaller than normal, and their quality as food for the larvae of E. autumnata (as well as for several sawfly species) is lower than after a warm summer (Senn, Hanhimäki & Haukioja 1992; Hanhimäki, Senn & Haukioja 1995). During the present study, summer temperatures were close to average, except for the summer of 1993 (Kuukausikatsaus Suomen ilmastoon, June–September 1993–95), which was cold and may thus have contributed to the low foliage quality for E. autumnata in 1994. However, the differences in adult size between 1994 and other years do not indicate any strong effects of weather.

A more likely explanation is asynchrony between phenologies of larvae and birches in spring 1994. Differences in average annual temperature and in the temporal accumulation of the temperature sum may affect the developmental rates of leaves and larvae differently, thus either fostering or deterring developmental synchrony between larvae and leaves (Ayres 1993; but see Neuvonen et al. 1996). Field observations in the present study indicated poor synchrony of larval hatching with leafing in 1994 (Kaitaniemi et al. 1997b), but its real effects are hard to prove. Trees at four outbreak sites were inspected at the same phenological stage in both 1993 and 1994; in the former year numerous larvae were present in all trees at the time of inspection, but in the latter year larvae were observed only occasionally, even at site no. 1, which had its peak density in 1994. Delayed hatching in relation to leaf development results in poor performance and small size of E. autumnata (Ayres & MacLean 1987; Kaitaniemi et al. 1997b). Furthermore, asynchrony may have increased larval parasitism in 1994, because it prolongs the period during which the larvae are vulnerable to parasitoids (Kaitaniemi et al. 1999).

In conclusion, the study reported here indicated only a weak DIR response in birches during and after the recent outbreak of E. autumnata in the central part of Finnish Lapland. Delayed induced resistance was not efficient enough to be the main reason for the termination of the outbreak, and it seemed unlikely that presence of DIR was required at all to terminate the outbreak. Whether this was a characteristic of the 1990s outbreak alone remains unclear, but Bylund's (1995) work with E. autumnata & Baltensweiler's (1993) work with Zeiraphera diniana indicate that different terminating agents may be important for different individual outbreak peaks, even within the same system. Even if plant-related traits are important for the outbreak, the weather or other environmental conditions may largely modify their extent and introduce variation into the behaviour of the system.


We thank Erkki Korpimäki, Seppo Neuvonen and Tero Klemola for their comments on earlier versions of this manuscript. Susanna Haapala, Lauri Kapari, Tero Klemola and Sari Mikkonen assisted with the field and laboratory work. Miia Tanhuanpää helped with the statistical analyses. Special thanks to Ellen Valle for checking our English. This study was financed by the Academy of Finland (e.g. grant no. 38425), and by the Jenny and Antti Wihuri Foundation.

Received 22 December 1997; revision received 3 July 1998