Cognitive ability measures: raw values

In all three cognitive ability measures (mental speed, fluid intelligence, crystallized intelligence) and in all three age groups (preschool, primary school, secondary school) Costa Rican children achieved lower results than Austrian children (see Table 1). The average difference across all three abilities and all three age groups is d=1.13 standard deviations. The difference is smaller in fluid compared with crystallized intelligence (d _Fl=1.08 vs d _Cr=1.24), but it is also large in mental speed (d _MS=1.21). The difference is smaller in preschool and primary school than in secondary school (d _Pre=0.78, d _Prim=0.77, d _Sec=1.84). The country pattern is by and large independent of the tests used to assess each ability construct. Within speed, preschool CPM-time vs school ZVT: d _CPMt=0.64 vs d _ZVT=1.51. Within fluid intelligence, preschool CPM, primary SPM and secondary APM: d _CPM=0.91, d _SPM=0.72, d _APM=1.60. Within crystallized intelligence: d _Read=1.20 vs d _Maths=1.27.

However, the differences increase with age for each of the three ability constructs: mental speed, d _Pre=0.64, d _Prim=0.92, d _Sec=2.09; fluid intelligence, d _Pre=0.91, d _Prim=0.72 and d _Sec=1.60; crystallized intelligence, d _Prim=0.66 and d _Sec=1.81. There is no indication that the secondary school samples are differently selected (as ‘Austrian elite’ or ‘Costa Rican sub-standard’).

Differences between the two countries increase when they are compared at the level of the g-factor: in primary school from d _Prim=0.77 to d _gPrim=0.95; in secondary school from d _Sec=1.84 to d _gSec=2.81. The Costa Rican samples contain more over-aged children (see Table 1, in kindergarten 7-year-old children, in 4th grade 12-year-old pupils, in the 15- and 16-year-old secondary school group 17- and 20-year-old students). There are also some ‘too young’ 14-year-old students. Age streaming seems to work less well in the Costa Rican educational system; or in Costa Rica more children with delayed development are kept in regular schools, perhaps due to integration policies or for lack of special schools for mentally challenged children.

The reliability is higher in Costa Rica than in Austria for the CPM (3 sets, Cronbach's α _CR=0.74 vs α _Au=0.66), the SPM (5 sets, Cronbach's α _CR=0.84 vs α _Au=0.72) and the APM (2 sets, Cronbach's α _CR=0.48 vs α _Au=0.38). The value for APM is rather low, there are only two subtests and the first twelve-item scale is more intended as a warm-up. The measures of crystallized intelligence in 4th grade have slightly higher reliability in the Austrian sample (PIRLS-Reading and TIMSS-Mathematics, Cronbach's α _CR=0.63 vs α _Au=0.67), similar to PISA-Reading and PISA-Mathematics in secondary school (Cronbach's α _CR=0.60 vs α _Au=0.64). (Unfortunately there are no ZVT results at the level of the four subtests, nor item-level results in the Ravens and student assessment tests.) There is no indication for generally lower reliability of the cognitive tests in the Costa Rican sample.

In both countries there is no correlation between CPM processing time and CPM solved tasks in preschool: r _CR=−0.01 vs r _Au=−0.07. The ‘mental speed’ substitute in kindergarten has low validity, possibly because of a speed–accuracy trade-off. Within primary school the correlation between mental speed and fluid and crystallized intelligence is higher in Costa Rica (Costa Rica: r _MS-Fl=0.41, r _MS-Cr=0.42, r _Fl-Cr=0.57; Austria: r _MS-Fl=0.29, r _MS-Cr=0.34, r _Fl-Cr=0.42). Also, in secondary school the correlation between mental speed and fluid and crystallized intelligence is on average higher in Costa Rica (Costa Rica: r _MS-Fl=0.29, r _MS-Cr=0.42, r _Fl-Cr=0.54; Austria: r _MS-Fl=0.01, r _MS-Cr=−0.04, r _Fl-Cr=0.54). Because of the higher correlations between tests the average g-factor loadings and variance explained by the g-factor are higher in Costa Rica: in primary school, λ _CR=0.76 vs λ _Au=0.72, explaining 57.30% vs 52.11%; in secondary school, λ _CR=0.74 vs λ _Au=0.59, explaining 54.55% vs 48.78%. The pattern corresponds to Spearman's law of diminishing returns, which says that g-loadings are lower at higher ability levels (Spearman, Reference Spearman1932; Jensen, Reference Jensen1998; Coyle & Rindermann, Reference Coyle and Rindermann2013). The pattern also underscores the validity of the Costa Rican ability results.

Cognitive ability measures: IQ values

In the first standardization procedure the results were standardized by defining the Austrian sample mean as M=99.28 IQ points (in Greenwich norm) and SD=15 (see Table 2). The total Costa Rican sample average is 82.38 IQ points. Differences between speed, fluid intelligence and crystallized intelligence are negligible (M _MS=81.00, M _Fl=83.09, M _Cr=80.75). Also the reading and mathematics results are similar (M _Read=81.33, M _Maths=80.17). The results in secondary school are well below the results of preschool and primary school (M _Pre=87.60, M _Prim=87.80 and M _Sec=71.73). Compared with other cognitive ability studies in Costa Rica (PISA 2009: IQ 88 for students, IQ 84 total youth corrected, IMO IQ 85) and with the Latin American average (student assessment: IQ 80, total youth corrected IQ 76, IMO IQ 80, psychometric corrected IQ 81) the results are similar.

Table 2.

IQ cognitive ability results, Costa Rican and Austrian children, 2008 (standardization based on Austrian results, defined as Greenwich 99.28 IQ points and SD=15)

In the second standardization procedure the results for ZVT and the Ravens were standardized using published test norms and, if necessary, a Greenwich adaptation and Flynn correction (see Table 3). For the ZVT, a ‘standard’ Flynn effect correction (−10.85 IQ points) would have produced an Austrian mean of 88.9 IQ points, clearly below the expected value of 100 or 99. But why should Austrian pupils differ so much in mental speed from German ones, when both countries obtain very similar results in psychometric and student assessment studies (in a current standardization, Austria IQ 99.28, Germany IQ 99.13)? As Silverman (Reference Silverman2010) has shown there is no indication for any secular rise in mental speed during the 20th century, and reaction times have actually become worse, implying an anti-Flynn effect. Therefore no Flynn effect correction was applied to the mental speed measure ZVT. This leads to an average IQ of 99.75 points for mental speed in the Austrian sample, which is compatible with the German standardization and with the results for fluid and crystallized intelligence. The Costa Rican average mental speed IQ of 78.33 is still 3 IQ points below the Costa Rican average derived from the standardization based on Austrian sample values (Table 2).

Table 3.

IQ cognitive ability results, Costa Rican and Austrian children, 2008 (standardization based on published test norms, Greenwich-adapted and Flynn-corrected)

Calculation of the Costa Rican values depends not only on the means of norms, derived either from this Austrian sample or from published test norms and their corrections, but also on the distributions in this Austrian sample or in the standardization samples, and on assumptions about the normality of latent ability distributions as opposed to measured scores. The ZVT standard deviations are larger in this Austrian sample than in the standardization sample (SD=17.31 vs 15.00), meaning the ZVT standardization sample was more homogeneous. Based on the properties of our sample, it is not plausible to assume that it is more heterogeneous than a representative German sample. (1) In the Austrian secondary school sample there were no students attending the high-ability Gymnasiums, and there were no immigrants. (2) Mentally handicapped children are missing. (3) The Austrian sample was recruited in small towns, not in the city of Vienna with its large cognitive elite and many low-scoring labourers and immigrants. (4) Flynn effects tend to be larger in low-ability groups than in high-ability groups leading to a more homogeneous age cohort. In correspondence with this the Raven results in the Austrian sample were more homogeneous than the standardization samples in the Raven manual (SD=12.21 vs 15.00).

These considerations led to a correction of the ZVT standard deviation. The Raven SD of around 12.21 based on this sample information is more plausible and additionally highly robust across age groups (SD_Pre=12.40, SD_Prim=12.01, SD_Sec=12.21). However, comparing the three cognitive ability measures speed, fluid intelligence and crystallized intelligence in Table 2 reveals that differences between countries are smaller on tests of fluid intelligence than on tests of speed or crystallized intelligence. Therefore it was decided not to influence the cross-country differences by using different standard deviations. One common standard deviation slightly above the Ravens' empirical mean but well below the ZVT mean was used: SD=13.00. This SD of 13 was also used for the student assessment comparisons.

Finally there are large differences across age. In preschool (using CPM) the results in both countries are higher than in primary school (using SPM), and in both countries the primary school results in the Raven are higher than the secondary school results (Table 3). This pattern for both countries is not plausible. The ZVT results indicate an average level in the Austrian secondary school sample. There is no indication that the preschool samples were elite samples. Maybe the Flynn correction was too small. The secondary school APM results (with Flynn correction) are far lower than the ZVT results (without Flynn correction). Flynn effects are frequently higher for younger children than for older youth and adults in countries as diverse as Britain (Flynn, Reference Flynn2009; Lynn, Reference Lynn2009, Table 3), USA (Rindermann & Thompson, Reference Rindermann and Thompson2013) and Taiwan (Chen et al., Reference Chen, Liao, Chen, Chen and Lynn2013). Thus the secondary age group Flynn correction was reduced from dec=2.0 to 0.5 points/decade, subtracting 1.45 instead of 5.80 points from the APM score. And a conservative downward correction of 3 points was applied to the preschool results of both countries, bringing the Austrian IQ to 103. The primary school sample was corrected upwards by 3 points, assuming that the Flynn correction was too strong or both samples were negatively selected. The results of ZVT and SPM are now similar.

All these corrections reduce the age-group differences, but do not abolish them. These corrections have their uncertainties, but there are no recent test standardizations in Great Britain and in this study in Austria and Costa Rica there are not the usually ‘large-scale’ study samples. Adjusting scores based on theoretical assumptions and empirical results leads to more valid outcomes.

Rather than following the above procedures, it is possible to suppose Flynn effects only for younger but not older students on the Raven tests (referring to Flynn, Reference Flynn2009) and on school-related tests (referring to TIMSS studies, which found improvement over time for 4th graders but not 8th graders; Mullis et al., Reference Mullis, Martin, Foy and Arora2012, pp. 51–64; Martin et al., Reference Martin, Mullis, Foy and Stanco2012, pp. 48–62). As long as there are no total current surveys in Great Britain and Costa Rica sample results cannot be used as unbiased estimates and therefore will never be free of considerations and corrections based on theory and empirical findings in achieving sound results. The authors also think that it is important to show the existing uncertainties and possible ways of dealing with them when researchers compare test scores across countries. This is common practice in economics, e.g. for comparing productivity, income and wealth, but is less commonly done in cross-cultural cognitive ability research.

The results of the third standardization procedure are presented in Table 4. Average cognitive ability levels across ability constructs, tests and age groups ‘are’ M _Au=99.20 and M _CR=87.33 IQ points. ‘Are’ is put in quotation marks to emphasize that these values depend on assumptions and corrections. Differences increase with age, from preschool d _IQ=7.45 and primary school d _IQ=8.34 to secondary school d _IQ=19.84. This general pattern could be observed with all scaling procedures. Differences are larger in crystallized intelligence than in fluid intelligence and mental speed, a pattern also observable in raw value comparisons and standardizations based on the Austrian sample (Tables 1 and 2). When g-factor scores (based on normed scales) are compared, differences between the countries increase from d _Prim=8.34 to d _Prim=13.27 IQ (standardized ds multiplied by 15; by 13: 11.50 IQ) in primary school and from d _Sec=19.84 to d _Sec=39.29 (multiplied by 13: 34.05) in secondary school.

Table 4.

IQ cognitive ability results, Costa Rican and Austrian children, 2008 (standardization based on norms, Austrian results and theoretical assumptions)

Sex/gender differences are rather unsystematic across ages. However, as in other studies (e.g. Flores-Mendoza et al., Reference Flores-Mendoza, Widaman, Rindermann, Primi, Mansur-Alves and Couto Pena2013), girls have higher mental speed (M _♂=93.21 and M _♀=95.48 IQ points, SD=11.97, same pattern in Austria and Costa Rica) whereas boys show higher fluid intelligence (M _♂=94.25 and M _♀=91.97 IQ points, SD=13.67, same pattern in Austria and Costa Rica). In crystallized intelligence there is no robust pattern. Girls very slightly outperform boys in reading literacy (M _♂=90.38 and M _♀=90.71 IQ points, SD=13.72), and boys outperform girls in mathematics (M _♂=92.27 and M _♀=88.24 IQ points, SD=12.65).

Cognitive ability estimation for total Costa Rican youth and society

Finally, we tried to estimate ability values for total Costa Rican youth and society. Some children and youth will not attend preschool and secondary school, but because primary attendance is mandatory it was assumed that the entire cohort is in primary school (CINDE, 2011). There are no private schools in the sample: these are attended mainly by the children of wealthier and better-educated parents, but only 8–10% of elementary and secondary school students are enrolled in private schools (CINDE, 2011; see also www.indexmundi.com/facts/costa-rica/school-enrollment). These two effects (omitting weaker youth out of school and smarter youth in private schools) tend to cancel each other out. Because no deviation larger than ±1 IQ point was expected, the student result was accepted as an estimate for the entire age cohort (IQ 87).

However, about 4% of Costa Rican adults are illiterate (CINDE, 2011), and the average level of adult schooling is lower than among youth. In 2000 only 91% of adults older than 24 years had ever attended school, and average length of school attendance for adults above 24 years was 6.01 years (compared with Austria: 8.80 years, UK: 9.35 years; Barro & Lee, Reference Barro and Lee2009). It is estimated that today's Cost Rican youth have an average of around 10 years of education (100% 6 years elementary and 91% 5 years secondary minus a dropout rate; www.tradingeconomics.com/costa-rica/school-enrollment-secondary-percent-gross-wb-data.html; 4th January 2012). It is assumed that in Costa Rica, being an ‘emerging’ country with most likely lower quality of schooling, schooling-related IQ gains are somewhat smaller (e.g. Van de Vijver & Brouwers, Reference Van de Vijver and Brouwers2009) than the usual 3 IQ points per year of school attendance (Winship & Korenman, Reference Winship, Korenman, Devlin, Fienberg, Resnick and Roeder1997). Additionally, school-related gains are generally smaller at older ages (Arum & Roksa, Reference Arum and Roksa2011). Therefore an IQ gain of 2 points per school year was estimated. This estimate is supported by the age-related increases in performance gaps in the sample. Thus 8 IQ points from the youth estimate of IQ 87 were subtracted (10 vs 6 years school attendance, 4 years×2 IQ points), resulting in an estimated average adult IQ of 79.

Some may have acquired more education as adults, and experiences other than schooling can raise intelligence. On the other hand, general developmental conditions in the past were almost certainly worse: poorer nutrition and health, no computers and related gadgets etc. For the entire Costa Rican population (in 2010: 0- to 19-year-old group around 34%, 20- to 49-year-old group 47%, 50- to 99-year-old group 19%; DESA, 2010), this results in an estimated average IQ of 82 (0.34×IQ 87+0.66×IQ 79). In a second version only the difference compared with Austria was used, the benchmark country: Costa Rican adults have on average 2.79 years less education (6.01 vs 8.80 years). This translates into an estimated adult IQ of 81 (IQ 87−(2.79×2 IQ points)). Using the same formula as above this results in an IQ of 83 for the total population (0.34×IQ 87+0.66×IQ 81). The last approach has the advantage that it considers also a lower adult educational level in Austria (i.e. considering the worldwide rise of education in calculating IQs).

Other measures

Height is a general indicator of standard of living (Komlos & Kriwy, Reference Komlos and Kriwy2003). In this study, Austrians are 6.80 cm taller than Costa Ricans (d=1.07, 16.04 HQ (height quotient) when scaled the same way as IQ, see Table 5). The difference is larger in secondary school (10.88 cm, d=1.44, 21.61 HQ) than in preschool (5.27 cm, d=1.08, 16.22 HQ) and primary school (4.26 cm, d=0.69, 10.28 HQ). The height differences are not due to undernutrition. The body mass index (BMI) is higher in Costa Rica (19.35 kg/m²) than in Austria (17.59 kg/m²). The difference in BMI units is −1.76 (d=−0.89, −13.37 BQ).

Table 5.

Height differences, Costa Rican and Austrian children, 2008

d-calculation: Austrian SD as reference sample; HQ, Height Quotient (similar to IQ) converted from d by multiplying by 15; means in d and HQ across ages (not age-groups) weighted for N.

Table 6 lists correlations of family and child background variables with country (positive correlation: higher values in Austria), with cognitive ability mean (speed, fluid and crystallized averaged) and with cognitive ability mean when country is partialled out. Parental educational level (school and professional degrees of both parents averaged, α=0.83) is similar in both countries, but robustly correlated with children's ability (r _p=0.28). One degree higher (e.g. from primary to secondary or from high school degree to university degree) corresponds to 1.96 IQ points (not corrected for country). Austrian families have more books (on average 208 vs 86 in Costa Rica), and 166 additional books add 1.00 IQ point (uncorrected). Austrian parents read more frequently to their children (in Costa Rica on average twice a week, in Austria four times). This corresponds to 2.41 IQ points. Children in Costa Rica read twice a week and in Austria three times; this corresponds to 2.11 IQ points. Common leisure time spent on Bildung (education and culture) is an index variable consisting of playing brain-teaser games, attending museums and theatres, playing an instrument and doing handcraft together with parents (Cronbach's α=0.60). This index is half a standard deviation higher in Austria (d=0.52). One standard deviation adds 1.73 IQ points.

Table 6.

Correlations of possible cognitive ability determinants with country and ability, Costa Rican and Austrian children, 2008

p=significance, r(country)=correlation with country (1 Costa Rica, 2 Austria, positive coefficients mean higher values in Austria); r(IQ)=correlation with cognitive ability sum (mental speed, fluid and crystallized intelligence), r _p(IQ, country)=correlation with cognitive ability sum, country partialled out.

In Costa Rica 69% of primary and secondary school children have attended preschool, and in Austria 95%. Attending preschool adds 10.40 IQ points (not corrected for country, corrected 5.76 points). Parental income (family income per month in Euro) is seven times higher in Austria than in Costa Rica (2204 vs 340 Euro); 250 Euro correspond to 1 IQ point (not corrected for country, corrected 2002 Euro for 1 IQ point). More children in Austria have their own desk at home (94% vs 63%). A desk contributes 7.82 IQ points (not corrected for country, corrected 1.82 points).

There is a small cross-country difference in vaccinations and periodical consultation of pediatricians (index-variable). Receiving all necessary vaccinations and regular pediatric check-ups corresponds to 8.38 IQ points. Parental participation in survey (filled questionnaire was returned to the test administrator by child) was better in Austria (90% vs 66%) and is associated with 8.90 IQ points (corrected for country 5.60 points). Discipline in school (no disrupting of instruction, not skipping class) is slightly higher in Austria: ‘sometimes’ instead of ‘frequently’ or ‘never’ instead of ‘sometimes’ is associated with 2.44 IQ points (uncorrected). Children in Costa Rica watch TV an average of 146 minutes per day; in Austria 98 minutes. One hour subtracts 2.28 IQ points (corrected for country 0.66 points). Of course, these variables are correlated among themselves and with country. Assumed causal effects and their theoretical meanings need clarification.

Attempt at a (causal) path analysis of national differences

Is it possible to explain the 12-point IQ gap between Costa Rica and Austria? For an explanation a convincing theory is needed that is compatible with results from earlier empirical research and that is supported by robust associations within and across countries. A model was developed (see Fig. 1, model has a good fit) starting with country and parental educational level. Country is an atheoretical variable that may represent aspects of culture, social and physical environment and genes. Country itself has no explanative value. Parental educational level stands for home environment, parenting styles and genes (Hart & Risley, Reference Hart and Risley1995). Ideally, it would be possible to reduce the direct (but atheoretical) effect of country on intelligence to zero by including as mediators all those variables that could create differences between countries and which are themselves influenced by ‘country’ (e.g. as culture or wealth). It was assumed that both country and parental education influence the environmental conditions that are important for children's cognitive development: higher parental income (β _{Country→PIn}=0.79, β _{Education→PIn}=0.19), number of books, children having their own desk, time parents spend with children on Bildung activities and preschool attendance. For all these hypothesized mediators the macro-social country effect is stronger than the individual parental effect. People shape their environment, but are constrained heavily by macrosocial conditions. One example: being skilled or not (e.g. as craftsperson or physician) will usually raise income everywhere, but being born in Western Europe vs Central America or in the 20th compared with the 19th century has a much larger effect.

Fig. 1.

Model for cross-country differences between Costa Rica and Austria (N _max=385, FIML, CFI=0.95, SRMR=0.04; C → means effect of country; E → means effect of parental education; positive effect of country means higher value in Austria; correlations are in parentheses).

Country and parental education have independent direct effects on children's cognitive ability (β _Co→CA=0.40, β _PEd→CA=0.23). Again the country effect is larger. Country accounts for 10.47 IQ points, one step higher in parental educational degree for 1.87 points. Many effects are indirect in the sense that mediating variables can be identified: in Austria more parents read more frequently to their children, and this appears to stimulate children's own reading either through modelling, creation of a habit or even through shared ‘genes for reading’. Frequent reading by the child can further the child's cognitive ability by training of thinking processes and acquisition of vocabulary and knowledge (Protzko et al., Reference Protzko, Aronson and Blair2013), although the association can also be caused by pleiotropic genes that influence both learning ability and the propensity for reading (β _OR→CA=0.15, 1.43 IQ points). Number of books, indicating environmental opportunities for learning and cognitive training, is an intervening variable (Rindermann et al., Reference Rindermann, Michou and Thompson2011). Also children who have their own desk have a better opportunity to learn (β _OD→CA=0.03, 0.78 IQ points). Parents who spend time with their children attending museums or theatres are also likely to read more to their children, and the children spend less time watching TV (β _TV→CA=−0.06, −0.67 IQ). Usually, watching TV is more a distraction than an element of Bildung (education and culture). Time for TV watching is likely to compete with time for reading, doing homework or other cognitively more stimulating activities. Therefore it can potentially have a negative impact on cognitive development.

Better educated parents care more for the health of their children through vaccinations and pediatric check-ups. This will further healthy development and thereby cognitive development (β _VP→CA=0.06, 3.99 IQ points). Another important variable is preschool attendance (β _Pre→CA=0.14, 4.70 IQ points). Early educational interventions, from special programmes to regular kindergarten, have been shown to have positive impacts on cognitive and personality development (Pianta et al., Reference Pianta, Barnett, Burchinal and Thornburg2009; Rindermann & Baumeister, Reference Rindermann and Baumeister2012). The positive impact of kindergarten attendance on medical care may have two causes. First, preschool educators can ‘diagnose’ sick children and initiate steps for medical referral. Second, preschool educators can influence parents' health care behaviour. Like other hypothesized mediators, these variables are also indicators for more general conditions, such as general interest of parents in the well-being of their children. In a simpler model (not displayed) considering only parental education, income and vaccination and pediatric visits, parental education has a stronger effect on health care than income (β _PEd→VP=0.17, r=0.19, β _PIn→VP=0.09, r=0.13). It's an ability effect, not a wealth effect (Gottfredson, Reference Gottfredson2004).

Finally, there is an important suppressor effect: parental income. In spite of very large country differences, this has no positive effect on children's cognitive development (β _PIn→CA=−0.10). This challenges any interpretation of cross-country cognitive differences as wealth difference or poverty effects. Some effects are rather small (β<0.10), but their ‘effects’ in the IQ metric are not negligible. These effects should be checked for generalizability, e.g. with data from the coming PISA surveys in Costa Rica. Their causality should be explored in experimental training studies.