DRAFT – FOR COMMENT, NOT QUOTATION Measures and determinants of non-cognitive ability across seven English-speaking countries Dr. Nicholas Biddle – Research School of Social Sciences, Australian National University, Canberra, Australia Sarah Ball – Research School of Social Sciences, Australian National University, Canberra, Australia Samuel Weldeegzie, – Crawford School of Public Policy and Research School of Social Sciences, Australian National University, Canberra, Australia [email protected] Centre for Aboriginal Economic Policy Research Copland Building #24 Australian National University Canberra, ACT, 2601 Australia Abstract While non-cognitive skills such as self-control and perseverance are found to be strong predictors of success, little is known about how these skill sets are distributed across countries and what determines them in a given country. After developing an index for such skills using principal component analysis from a range of questions, this paper uses the most recent data from PISA 2012 to examine what factors are associated with cognitive and noncognitive abilities in seven English-speaking countries. However, we still need further research on identifying exactly which skills are most important and improve measurement thereof. Keywords: Non-cognitive ability, Programme of International Student Assessment, Education policy, Educational attainment DRAFT – FOR COMMENT, NOT QUOTATION Measures and determinants of non-cognitive ability across seven English-speaking countries 1. Introduction Education plays a key role in alleviating the negative outcomes associated with poverty and disadvantage. There is an extensive literature on measuring returns to education with experimental, quasi-experimental and observational data showing that those individuals, communities and countries that have relatively high levels of education tend to have better outcomes across a range of economic and other wellbeing measures (Psacharopoulos and Patrinos, 2004, Carneiro, Heckman et al. 2010). Other research has shown that the education levels of parents has a significant and substantial effect on the education outcomes of their children (Black and Devereux 2011). Policy in most developed countries reflects this with considerable public investment in early childhood, school and post-school education. However, this investment has tended to focus on school attainment and measures of cognitive ability such as test scores in Mathematics, Literacy and Science. In the United States (US), one of the major policy focuses over the last decade has been No Child Left Behind (NCLB), which has been based on, and assessed against standardised student assessments of mathematics and literacy (Dee and Jacob 2011). Similarly, a major policy driver in Australia is NAPLAN, or the National Assessment Program – Literacy and Numeracy. These sets of tests, administered to students across Australia in Years 3, 5, 7 and 9, make up a large focus of the My School1 website in which all schools in Australia are ranked. This information has been used extensively in arguing for changes to school funding formulas. Similarly, when countries are ranked internationally in terms of education performance, a similar set of tests are used, administered as part of the OECD’s Programme for International Student Assessment (PISA) for 15 year olds. Like NAPLAN, the PISA tests focus on mathematics and reading ability, but also extend to science scores. While the focus of the PISA is on comparisons of literacy, numeracy and science ability, significant personal and attitudinal data is also collected through a background questionnaire. This paper argues that while cognitive ability is important, skills like self-control, persistence and determination have also been shown to have a major impact on educational outcomes and later measures of wellbeing and are therefore deserving of a much greater policy and research focus. This set of skills is commonly referred to as non-cognitive ability and has been shown to improve the likelihood of school completion (Heckman 2008, Barón and Cobb-Clark 2010), employment (Mischel, Shoda et al. 1989, Heckman and Masterov 2007), and even reduce crime and single parenthood (Heckman 2006, Heckman and Kautz 2013). While the PISA results are used to measure cognitive ability across countries, we know very little about how non-cognitive ability is distributed internationally, and what the within- 1 http://www.myschool.edu.au/ DRAFT – FOR COMMENT, NOT QUOTATION country determinants are. This paper aims to fill this gap. For reasons outlined in the methodology section, we focus our analysis on seven English speaking countries – Australia, Canada, Ireland, New Zealand, Scotland, UK (excluding Scotland), and the USA. We begin with a discussion of previous research on non-cognitive ability (in Section 2), followed by a discussion of issues related to measurement (in Section 3). We then discuss our methodology (in Section 4), followed by a set of descriptive statistics (Section 5) and more detailed modelling (Section 6). The final section in the paper provides a summary and some concluding comments. 2. Previous research The role of non-cognitive skills in early life and education is supported by a growing body of research. Research on non-cognitive ability rose to prominence (at least within the field of economics) through the work of James Heckman. With colleagues, Heckman has undertaken a number of significant studies that highlight the role of non-cognitive ability on a variety of socioeconomic outcomes. Heckman and Rubinstein (2001) for example, analysed students who completed the General Education Development (GED) program, which certifies that students have equivalent cognitive ability to a high school graduate. The study found that those who complete the GED do not have better outcomes than their peers who do not complete the GED. The authors argue that this is because GED students lack the noncognitive skills of their school-completing peers (Heckman and Rubinstein 2001). Heckman’s work on the long-term benefits of investment in early childhood is also notable (Heckman 2000, Heckman 2006, Heckman and Masterov 2007, Heckman, Moon et al. 2010). A key example is his work on the Perry Pre-school Programme. The programme was carried out from 1962-67 and provided high quality pre-school education to 128 African-American children. Children who participated in the evaluation of the programme were followed up at 27 and 40 years of age. Considered somewhat of a failed experiment, by 8-9 years of age the IQ scores of participants were no better than the control. However, in the later follow up children who had participated showed higher rates of school completion, were more likely to be employed and were less likely to have spent time on welfare. In searching for why this might have occurred, Heckman and colleagues decided to look at data related to personal behaviour which had been collected but never analysed. This research showed an improvement in non-cognitive factors, like self-control, which could be attributed to as much as 65% of the total benefit of the intervention (Heckman and Masterov 2007, Tough 2012). The work of Walter Mischel is also significant when considering non-cognitive skills. The famous Stanford Marshmallow test led by Mischel has contributed much to our understanding of delayed gratification and self-control and its impact on life outcomes. Mischel conducted the test in the late 1960s at Stanford University using a group of preschoolers. A child is asked to choose between one treat now (a marshmallow for example) or to wait a short time for a two treats later. It is essentially a test of self-control, willpower and delayed gratification (Mischel, Shoda et al. 1989). Mischel tracked many of these children for decades, collecting data on each child’s education, health and other factors. He found that those who had resisted temptation on average had higher academic scores, achieved higher DRAFT – FOR COMMENT, NOT QUOTATION educational degrees, earned more money and even had a lower body mass index (Mischel 2014) . A recent paper released by the Brookings Institution titled The Character Factor: Measures and Impact of Drive and Prudence evaluated existing datasets from the US in order to find reliable measures of non-cognitive skills such as self-control, delayed gratification and persistence (Reeves, Venator et al. 2014). The authors then used one of these measures (a hyperactivity scale) to estimate its influence in the early and middle childhood years for educational and other outcomes. Their findings show a significant relationship between drive and prudence and higher educational attainment even when controlling for other factors. They also show that children from disadvantaged backgrounds were more likely to perform poorly on the early measures of non-cognitive skill (Reeves, Venator et al. 2014). In the United Kingdom the Effective Pre-school, Primary and Secondary Education 3-16+ (EPPSE 3-16+) project is currently being used to explore the impact of pre-school on child development. This research also highlights the role of non-cognitive skills on developmental trajectories. The project is a longitudinal, mixed methods study, commissioned by the Department for Education. It has tracked 3172 students since 1997 across three phases of schooling (pre-school, primary and secondary school) and is now looking at their posteducational outcomes (Sammons, Sylva et al. 2014). In addition to academic and socialbehavioural outcomes from testing, grades and teacher ratings, the EPPSE 3-16+ included questionnaires on student disposition (including non-cognitive markers such as selfregulation). These questionnaires used self-reported measures and while acknowledging the issues of reliability the researchers state that, “there is a definite move towards both acknowledging the importance of student well-being and non-cognitive outcomes, and the development of robust measures to capture these aspects of young people's experiences.” (Sammons, Sylva et al. 2014). Research related to other comparable countries is somewhat more limited. In Australia, the most recent, relevant work has been done by Professor Deborah Cobb-Clark (Barón and Cobb-Clark 2010, Cobb-Clark and Tan 2011). In her 2010 paper with Juan D. Baron, CobbClark uses data collected for the Youth in Focus Project2, specifically focusing on the locus of control. She refers to literature which shows that those with an internal sense of control are likely to be more hardworking, motivated and persistent in problem solving (Gatz and Karel 1993, Coleman and DeLeire 2003). Cobb-Clark and Baron found that young people with an internal locus of control were more likely to complete school and on completion meet the requirements to enter university. The Youth in Focus Survey is about the experiences of young people in Australia. It asks questions about family background, living arrangements, education, work, relationships, income, health, spare time, and aspirations and attitudes. It also uses administrative data collected from the social security system. Two waves were conducted in 2006 and 2008. Young people aged 18 and one of their parents or carers were interviewed. The first wave included a sample of 4000 young people and 3900 parents or carers. 2 DRAFT – FOR COMMENT, NOT QUOTATION 3. Measurement Despite the research noted above, education policy debate appears to focus nearly exclusively on the development of cognitive skills. A key example of this can be seen in the recent review of the Australian Curriculum which stated that personal and social capability, critical and creative thinking, and ethical and intercultural understanding, should no longer be treated in a cross-curricular fashion and only be embedded in subjects and areas of learning where relevant (Department of Education 2014). One of the key reasons for government policy to focus on cognitive skills is the ability to standardise measures across the country, and internationally. Measuring non-cognitive skills is significantly more problematic and research shows that the complexity and breadth of noncognitive skills has impaired its use in research and policy (Reeves, Venator et al. 2014). There are several key reasons why non-cognitive measures have proven problematic. Firstly, when dealing with social and emotional traits it can be difficult to separate the findings from situational and contextual influences. Secondly, much research relies on self-reported measures which implicitly assume that respondents are capable of unbiased assessment of their personality. Finally, it can also be a challenge to ensure consistency of answers to a psychometric task over time or across observations (Hoover 2013, Reeves, Venator et al. 2014, West, Kraft et al. 2014). The overall risk of unreliable measures and issues of validity continue to plague more standardised approaches of measuring non-cognitive skill (Kyllonen 2005). Angela Duckworth however has done extensive work on the value of non-cognitive skills in predicting academic performance and her use of the ‘grit’ scale to measure determination and drive is a centrepiece in the current research (Duckworth and Seligman 2005, Duckworth, Quinn et al. 2012). The grit scale uses twelve brief statements, such as ‘I am a hard worker’ or ‘My interests change from year to year’, on which respondents must rate themselves on a 5 point scale. While simple, it was shown to be surprisingly effective at predicting later success (Duckworth, Peterson et al. 2007). A key example arose from work Duckworth and colleagues did with West Point Military Academy. The grit scale was used to test over 1200 cadets as they entered summer training and proved more effective than the Academy’s own extensive evaluation system at predicting which cadets would complete the training (Duckworth, Peterson et al. 2007). Duckworth and colleagues have done extensive work on the grit scale and significant research to show its effectiveness (Duckworth and Seligman 2005, Duckworth, Peterson et al. 2007, Borghans, Duckworth et al. 2008, Duckworth, Quinn et al. 2012, West, Kraft et al. 2014). However, they also openly admit that no measure is perfect (Duckworth, Quinn et al. 2011). The very extensiveness of the traits that can be considered and the issue of validating self-reported data, continue to make standardisation a major challenge (Reeves, Venator et al. 2014). One possible approach that has been suggested is the use of resources that have been developed by psychologists for evaluating non-cognitive skills. Some of these measures include the Big Five taxonomy (Borghans, Duckworth et al. 2008, Cobb-Clark and Tan DRAFT – FOR COMMENT, NOT QUOTATION 2011), the Mayer-Salovey-Caruso Emotional Intelligence Test (MSCEIT) (Kyllonen 2005) and a recent study considered using survey item response rates (Hitt, Trivitt et al. 2014). However these are yet to gain widespread acceptance for measuring the effectiveness of education policy (Heckman and Rubinstein 2001). 4. Methodology The most recent version of PISA3 provides an opportunity to address both the current gap in data and the standardisation issue. PISA is an international set of surveys that aims to test the skills and knowledge of students aged 15 years across a range of countries or regions within countries. Surveys are conducted every three years, with the most recent survey being undertaken in 2012 across 66 distinct countries or regions. A range of information is collected on the students, with a particular focus on standardised tests across math, reading and science. In addition though, a range of background information is collected. In addition to these measures of academic ability and outcomes, information is collected in PISA on other education-related measures, some of which fall under the general heading of noncognitive ability (Heckman and Mosso 2014) or what Reeves and colleagues (2014) have labelled drive or prudence. Non-cognitive ability is measured most explicitly by two sets of questions – perceived selfcontrol, and perseverance. While every student answered the questions on test scores, not everyone answered the questions on perceived self-control or perseverance. However, there is full information on 304,974 students for the former and 308,482 students for the later. This accounts about 63% of the total sample of students. In Appendix 3, we show that these students are significantly different in their cognitive ability (reflected in test scores) compared to those who didn’t reply to non-cognitive ability questions, with higher scores on average. In terms of perceived self-control, students were asked about the extent to which six specific characteristics described them well. Specifically, they were asked about whether in their view: • • • • • • They can succeed with enough effort; It is their choice they will be good at school; There are problems that prevent them from putting effort into school; Different teachers would make them try harder; They could perform well if they wanted; and They would perform poorly at school regardless of their effort. These measures are broadly related to the concept of self-efficacy. According to Julio Garcia and Geoffrey Cohen, the concept of self-efficacy has significant explanatory power with regards to differences in education outcomes. Children of equal ability will respond differently when challenged because; ‘…those who doubt their ability to succeed in school or students who believe that their level of intelligence is a fixed quality, are more likely than their peers to give 3 http://pisa2012.acer.edu.au/downloads.php DRAFT – FOR COMMENT, NOT QUOTATION up, persevere in ineffective strategies, experience negative emotions, and fail to return to their original performance level following failure’ (Garcia and Cohen 2012) Feeling that you have control over your environment is one thing. But exercising that control requires sustained effort. In terms of perseverance, students in PISA were asked whether: • • • • • They give up easily; They put off difficult problems; They remain interested in tasks that they start; They continue to work on tasks until perfection; and When confronted with a problem they do more than what is expected of them. Questionnaires such as these are complicated by their use of measures of self-perception, rather than actual character traits. They are also only able to capture a limited number of aspects of non-cognitive ability. Also, as the PISA questionnaire is administered as part of a cross-sectional survey it can only tell us what is associated with non-cognitive ability, not what it predicts. Nonetheless, they are useful proxies for the broad set of skills discussed earlier. For our analysis, we calculate an index of perceived self-control and perseverance that across the entire PISA sample had a mean of zero and standard deviation of one. We followed a four-step process separately for the six questions on perceived self-control, and the five questions on perseverance. That is, we: • • • • Undertake Principal Components Analysis (PCA) on the student responses; Calculate a preliminary index value for each individual based on the scoring coefficients; Calculate a mean and standard deviation for the whole sample using that preliminary index value; and For each individual, subtract the sample mean from each individual’s preliminary index value and divide it by the sample standard deviation. In order to make comparisons with standard measures of academic ability, we took an average of each student’s maths, reading and science scores. We then scaled in a similar way by subtracting the sample mean and dividing it by the standard deviation. In the first section of results (Section 5), we compare the measures of non-cognitive ability by country or region across the entire PISA sample. Given issues of language and culture, we then focus our comparisons across seven English speaking, developed countries – Australia, Canada, Ireland, New Zealand, Scotland, the UK (excluding Scotland) and the USA. The second section of results in the paper looks at the factors associated with the indexes of test scores and non-cognitive ability. Analysis is conducted first using a pooled sample across the seven English-speaking countries mentioned above (with a dummy variable for six of the seven countries). We then estimate the models separately for each of the seven countries. 5. Non-cognitive ability across the PISA sample There is significant and substantial variation across countries and regions for the two measures of non-cognitive ability. Keeping in mind that across the sample the variables have DRAFT – FOR COMMENT, NOT QUOTATION a mean of zero and a standard deviation of one, average perceived self-control ranges from a value of 0.48 in Costa Rica (highest non-cognitive ability on this measure) to -0.63 in MacaoChina (lowest ability). Perseverance, on the other hand, ranges from 0.59 in Kazakhstan to 0.81 in Japan. The next few figures summarise the spread across the rest of the countries. The first figure for each of the variables is for those countries with an average above the sample mean. The second figure for each of the variables is for those countries with average values below the sample mean. The ‘whiskers’ around the bars are for the 95% confidence interval. DRAFT – FOR COMMENT, NOT QUOTATION Figure 1 Countries with above average perceived self-control Figure 2 Countries with below average perceived self-control DRAFT – FOR COMMENT, NOT QUOTATION Figure 3 Countries with above average perseverance Figure 4 Countries with below average perseverance In addition to the very large spread across countries, the figures show that those countries that tend to do well on standardised test scores don’t necessarily do well on measures of noncognitive ability. Indeed, at the aggregate level, there is a negative correlation between test scores and the two measures (-0.17 for test scores and self-control and -0.51 for test scores and perseverance). This implies that countries with high average test scores have low average values for non-cognitive ability. This is a troubling finding for the measurement of noncognitive ability using such questionnaires. Given the research linked above, we would expect to find a positive correlation. There could be three reasons why this might be the case, all of which point to caution in how measures of non-cognitive ability should be used. The first issue is selection effect – those students who are doing well in test scores may be from relatively rich countries but are not answering the questions on non-cognitive items. Interestingly, the data allows us to test this. DRAFT – FOR COMMENT, NOT QUOTATION Comparing mean outcomes of test scores shows that those students with missing noncognitive data have significantly lower test scores than their counterparts (see Appendix 3). Therefore, we can’t conclude that, from a simple correlation observed, countries with high average test scores would have low average values for non-cognitive ability for the entire population. Another issue could be aggregation. The research summarised earlier looked at variation within countries. It may not be that non-cognitive ability predicts how well a country is doing, but how a person is positioned within a country. This is borne out by the data. When we looked at correlations across individuals, test scores were positively correlated with perceived self-control (with a value of 0.20) and to a lesser extent perseverance (0.05). A third issue in comparing across countries relates to language and cultural norms. Given the list of countries in the figures above, the questionnaires for PISA need to be translated into dozens of languages, with the potential for these language differences to impact on measured variation. There are also cultural norms that need to be taken into account, with the survey respondents taking into account, at least in part, what they think the right answer should be. We explore these issues further in the next section by looking at the individual level data within the seven English-speaking countries. 6. Variation in non-cognitive ability within countries We begin our analysis of individual data by analysing a pooled sample of students across the seven English-speaking, developed countries presented in Figure 5. Results are presented in full in Appendix 1, but are summarised in Figure 5 based on a linear regression analysis taking into account weights and sample design. Results are presented as the difference in predicted index values whilst holding other observed characteristics constant. Appendix 2 replicates the analysis within each of the seven countries separately. Most of the 16 explanatory variables summarised in Figure 5 are binary variables and are reasonably self-explanatory. Of the other two, the variable for parental education is measured as the number of years of education for the parent of the child that has the highest level of schooling. We present the association with an additional year of schooling. The final variable in the figure is an index of wealth of the student’s household created by the OECD, based on a range of questions on household possessions. We standardised this index to have a mean of zero and a standard deviation of one. Results presented in the figure for this variable represent the association with a one standard deviation increase in the index value. In essence, for all variables, bars to the right are those where a person with that particular characteristic has better outcomes than those in brackets, bars to the left are where that person has worse outcomes. Summary statistics for the three analyses are given below. What they show, and this needs to be kept in mind when interpreting results for individual coefficients, is that the standard demographic, early childhood, and family background characteristics explain a lot less of the variation in measures of non-cognitive ability compared to variation in test scores. • Test scores – Sample size 53,439, R-Squared 0.1423 • Perceived self-control – Sample size 34,198, R-Squared 0.0212 • Perseverance – Sample size 34,565, R-Squared 0.0337 DRAFT – FOR COMMENT, NOT QUOTATION DRAFT – FOR COMMENT, NOT QUOTATION Figure 5.Variation in index values within seven English-speaking developed countries. female born overseas speaks other than English attended preschool parent's schooling mother employed part-time mother not emplyoed father employed part-time father not employed wealth index -.3 -.2 -.1 0 .1 .2 Associated change in index values Test Persevere .3 Control Canada Ireland New Zealand UK Scotland USA -.5 -.4 -.3 -.2 -.1 0 .1 Associated change in index values Test Persevere .2 .3 Control Looking at the first section of the table, there were a number of individual and family-level variables that had a similar association with the two measures of non-cognitive ability as with the standardised test scores. In the seven English-speaking, developed countries, being born overseas, having relatively well-educated parents and living in a relatively wealthy household was associated with higher standardised test scores and higher measured non-cognitive DRAFT – FOR COMMENT, NOT QUOTATION ability. Having a father who was employed part-time (compared to having a father who was employed full-time) was associated with lower values. What is more interesting from a policy point of view though was those variables where the association differed, either in terms of direction or statistical significance. Females, for example, had higher standardised test scores and perceived self-control, but lower perseverance. This pattern was also found for those who attended preschool (early childhood education before the first year of full-time schooling), with the reverse found for those who speak a language other than English at home (lower test scores and self-control, but higher perseverance). The size of the associations were not as large, but low maternal employment (not employed or part-time employed) was found to be positively associated with test scores, negatively associated with perceived self-control, and not associated with perseverance. There were some differences in the country groupings once individual and family-level variables had been controlled for. Australia was included as the base case for no other reason than alphabetical and the fact that the authors were based at Australian universities. The more detailed modelling (figure 5) showed that Canada has higher test scores and measured noncognitive ability than Australia, whereas the US has lower test scores alongside the highest perceived control and perseverance. Interestingly though, the US was found to have higher test scores than the United Kingdom (including and excluding Scotland) once individual and other characteristics have been controlled for. The results for Ireland also changed somewhat, with higher measured test scores and non-cognitive ability once observable characteristics were held constant. 7. Discussion and concluding comments Angrist and Pischke (2010) have argued that ‘empirical microeconomics has experienced a credibility revolution, with a consequent increase in policy relevance and scientific impact’ and that ‘a clear-eyed focus on research design is at the heart of’ this revolution. If it hadn’t before, economics has clearly embraced experiments with a justifiable focus on causal inference. We are in strong support of this statement. There is, however, still a need for robust descriptive analysis that starts with associations in order to point the way towards carefully designed evaluations and theoretical innovations. Non-cognitive ability has the capacity to improve outcomes for all students over the lifespan by improving school completion rates and labour market outcomes. This is particularly important for those students who are already disadvantaged. These skills are also of importance in regard to equality of opportunity as there appears to be an achievement gap between wealthy and disadvantaged students. It is reasonably easy for policy makers to appreciate the role that literacy, numeracy and science knowledge will have on future outcomes and productivity. There is equally compelling evidence though that non-cognitive ability (also often labelled as character skills) strongly predicts future labour market, education and life success (Heckman and Kautz 2013). The aim of this paper was twofold. First, we use a PCA to develop an index for measures of non-cognitive abilities from a range of questions asked on how students perceive their perseverance and self-control. Second, we explore the variation across countries, and across individuals within selected countries in three measures – standardised test scores (averaged across Mathematics, Reading and Science), perceived self-control, and perseverance DRAFT – FOR COMMENT, NOT QUOTATION A number of important, policy relevant findings emerge. First, there is non-random, nonresponse to the questions on self-control and perseverance with those who did not answer the question having lower test scores than those who did. Cross-country analysis of noncognitive ability needs to reflect this. Second, for those who did answer the questions, there is a negative cross-country correlation between test scores and non-cognitive ability. There are potential cross-cultural and language reasons for this. However, the important point is that we cannot assume that education policy frameworks directed at reading, Mathematics and science will also optimise non-cognitive ability. Within seven English-speaking countries where cross-cultural and language issues are minimised (if not eliminated) there are important patterns in the variation of test scores and non-cognitive ability. Sex had a complex association with females doing relatively well in self-control and test scores, but relatively poorly in terms of perseverance. Many of the variables that we know are associated with test scores or measures of cognitive ability are also associated with measures of non-cognitive ability. In particular, both preschool attendance and parental education have a strong association with at least one of the measures of non-cognitive ability. We control for these student level characteristics in our analysis of variation across countries and find significant variation across the three measures that, importantly, do not necessarily move in the same direction. Canada and Ireland are the two countries in our analysis that do relatively well on all three variables. New Zealand does relatively poorly on all three measures. Australia does relatively well on test scores, but relatively poorly on the two measures of non-cognitive ability. The UK (excluding and including Scotland) do poorly on test scores but moderately on non-cognitive ability. Finally, the US does quite poorly on test scores but very well on measures of non-cognitive ability. There are no large-scale evaluations that we know of that would point to specific policies that are driving these differences. But, the descriptive analysis would suggest a need for such research. Ultimately, as economists and other social scientists increasingly recognise the importance of non-cognitive ability for future labour market and broader outcomes, surveys like PISA will help form the evidence-base for within- and across-country comparisons. At the same time, survey items included in PISA or similar will hopefully be used in a greater range of trials and evaluations. The analysis in this paper has shown that these variables have important information in them, but cannot be used unquestioningly. DRAFT – FOR COMMENT, NOT QUOTATION 5. Bibliography Barón, J. D. and D. A. Cobb-Clark (2010). "Are young people's educational outcomes linked to their sense of control?" Available at SSRN 1594792. Black, S. E. and P. J. Devereux (2011). "Recent developments in intergenerational mobility." Handbook of labor economics 4: 1487-1541. Borghans, L., A. L. Duckworth, J. J. Heckman and B. Ter Weel (2008). "The economics and psychology of personality traits." Journal of Human Resources 43(4): 972-1059. Carneiro, P., J. J. Heckman and E. J. Vytlacil (2010). Estimating marginal returns to education, National Bureau of Economic Research. Cobb-Clark, D. A. and M. Tan (2011). "Noncognitive skills, occupational attainment, and relative wages." Labour Economics 18(1): 1-13. Coleman, M. and T. DeLeire (2003). "An economic model of locus of control and the human capital investment decision." Journal of Human Resources 38(3): 701-721. Dee, T. S. and B. Jacob (2011). "The impact of No Child Left Behind on student achievement." Journal of Policy Analysis and Management 30(3): 418-446. Department of Education (2014). Review of the Australian Curriculum Final Report Canberra. Duckworth, A. L., C. Peterson, M. D. Matthews and D. R. Kelly (2007). "Grit: perseverance and passion for long-term goals." Journal of personality and social psychology 92(6): 1087. Duckworth, A. L., P. D. Quinn, D. R. Lynam, R. Loeber and M. Stouthamer-Loeber (2011). "Role of test motivation in intelligence testing." Proceedings of the National Academy of Sciences 108(19): 7716-7720. Duckworth, A. L., P. D. Quinn and E. Tsukayama (2012). "WhatNo Child Left Behind leaves behind: The roles of IQ and self-control in predicting standardized achievement test scores and report card grades." Journal of educational psychology 104(2): 439. Duckworth, A. L. and M. E. Seligman (2005). "Self-discipline outdoes IQ in predicting academic performance of adolescents." Psychological science 16(12): 939-944. Garcia, J. and G. Cohen (2012). A social psychological approach to educational intervention. Behavioral foundations of public policy. E. Shafir. Princeton, Princeton University Press: 329-347. Gatz, M. and M. J. Karel (1993). "Individual change in perceived control over 20 years." International Journal of Behavioral Development 16(2): 305-322. Heckman, J. J. (2000). "Policies to foster human capital." Research in economics 54(1): 3-56. Heckman, J. J. (2006). "Skill formation and the economics of investing in disadvantaged children." Science 312(5782): 1900-1902. Heckman, J. J. (2008). "Schools, skills, and synapses." Economic inquiry 46(3): 289-324. Heckman, J. J. and T. Kautz (2013). Fostering and measuring skills: Interventions that improve character and cognition, National Bureau of Economic Research. Heckman, J. J. and D. V. Masterov (2007). "The productivity argument for investing in young children." Applied Economic Perspectives and Policy 29(3): 446-493. DRAFT – FOR COMMENT, NOT QUOTATION Heckman, J. J., S. H. Moon, R. Pinto, P. A. Savelyev and A. Yavitz (2010). "The rate of return to the HighScope Perry Preschool Program." Journal of Public Economics 94(1): 114128. Heckman, J. J. and S. Mosso (2014). The Economics of Human Development and Social Mobility, National Bureau of Economic Research. Heckman, J. J. and Y. Rubinstein (2001). "The importance of noncognitive skills: Lessons from the GED testing program." American Economic Review: 145-149. Hitt, C., J. Trivitt and A. Cheng (2014). "When You Say Nothing at All: The Surprisingly Predictive Power of Student Effort on Surveys." Hoover, E. (2013). "Colleges Seek 'Noncognitive' Measures of Applicants; Admissions offices want to know about traits, like leadership, initiative, and grit, that the SAT doesn't test." The Chronicle of Higher Education 59(19). Kyllonen, P. C. (2005). The Case for Noncognitive Assessments. R&D Connections. E. T. S. R. a. Development. New Jersey, Princeton. Mischel, W. (2014). The Marshmallow Test: Mastering Self-Control, Little, Brown and Company Mischel, W., Y. Shoda and M. I. Rodriguez (1989). "Delay of gratification in children." Science 244(4907): 933-938. Psacharopoulos, G. and H. A. Patrinos* (2004). "Returns to investment in education: a further update." Education economics 12(2): 111-134. Reeves, R., J. Venator and K. Howard (2014). The Character Factor: Measures and Impact of Drive and Prudence. C. o. C. a. Familes, Brookings Institute. Sammons, P., K. Sylva, E. Melhuish, I. Siraj, B. Taggart, R. Smees and K. Toth (2014). "The Effective Pre-school, Primary and Secondary Education Project (EPPSE 3-16+) Students’ views of school in Key Stage 4 at age 16." Institute of Education, University of London/Department for Education http://www. ioe. ac. uk/research/153. html. Tough, P. (2012). How Children Succeed: Grit, Curiosity, and the Hidden Power of Character, Houghton Mifflin Harcourt. West, M. R., M. A. Kraft, A. S. Finn, R. Martin, A. L. Duckworth, C. F. Gabrieli and J. D. Gabrieli (2014). "Promise and Paradox: Measuring Students’ Non-cognitive Skills and the Impact of Schooling." DRAFT – FOR COMMENT, NOT QUOTATION Appendix 1: full results for comparisons across English speaking countries only: all sample VARIABLES Female Born Overseas Speaks other than English Grade level of student Attended Preschool Age started school Parent's level of schooling Mother employed part-time (ref: full time) Mother not employed Father employed part-time Father not employed Wealth (standard deviation) Canada (ref: Australia) Ireland New Zealand UK (excluding Scotland) Scotland USA Constant Observations R-squared (1) test scores (2) Self-control (3) perseverance 0.02** (0.010) 0.11*** (0.022) -0.14*** (0.016) 0.37*** (0.014) 0.18*** (0.023) 0.02*** (0.008) 0.07*** (0.003) 0.08*** 0.11*** (0.012) 0.05 (0.029) -0.02 (0.021) 0.07*** (0.017) 0.17*** (0.063) 0.00 (0.011) 0.03*** (0.003) -0.02 -0.05*** (0.013) 0.14*** (0.028) 0.18*** (0.027) 0.06*** (0.013) -0.03 (0.039) -0.03** (0.013) 0.03*** (0.004) 0.02 (0.013) 0.03** (0.014) -0.18*** (0.018) -0.14*** (0.016) 0.10*** (0.008) 0.10*** (0.011) 0.24*** (0.015) -0.26*** (0.020) -0.43*** (0.022) -0.46*** (0.024) -0.24*** (0.016) -4.52*** (0.143) 53,439 0.142 (0.026) -0.05** (0.019) -0.11*** (0.027) -0.10*** (0.026) 0.06*** (0.012) 0.10*** (0.015) 0.05** (0.021) -0.02 (0.021) 0.02 (0.020) 0.06** (0.024) 0.20*** (0.018) -1.27*** (0.196) 34,198 0.021 (0.022) 0.01 (0.015) -0.15*** (0.033) -0.00 (0.022) 0.09*** (0.009) 0.07*** (0.012) 0.07*** (0.017) -0.13*** (0.017) -0.02 (0.016) -0.02 (0.020) 0.28*** (0.013) -0.99*** (0.167) 34,565 0.034 DRAFT – FOR COMMENT, NOT QUOTATION Appendix 2: full results for each English speaking country: 2.1 Australia VARIABLES Female Born Overseas Speaks other than English Grade level of student Attended Preschool Age started school Parent's level of schooling Mother employed part-time Mother not employed Father employed part-time Father not employed Wealth-standard deviation Constant Observations R-squared (1) Test scores (2) Self-control (3) perseverance -0.01 (0.014) 0.02 (0.017) 0.15*** (0.023) 0.25*** (0.013) 0.30*** (0.028) -0.09*** (0.007) 0.13*** (0.003) 0.15*** (0.010) 0.02** (0.012) -0.06*** (0.020) -0.18*** (0.016) -0.05*** (0.007) -3.61*** (0.141) 0.01 (0.014) 0.03 (0.019) 0.25*** (0.024) 0.01 (0.013) 0.08** (0.039) -0.05*** (0.010) 0.05*** (0.004) 0.05*** (0.017) -0.00 (0.017) -0.09*** (0.026) -0.11*** (0.025) 0.01 (0.010) -0.70*** (0.167) -0.16*** (0.011) 0.19*** (0.020) 0.25*** (0.018) -0.02 (0.013) 0.14*** (0.028) -0.03*** (0.009) 0.06*** (0.004) 0.05*** (0.016) 0.10*** (0.014) -0.08*** (0.023) -0.10*** (0.024) 0.04*** (0.011) -0.65*** (0.163) 11,530 0.119 7,424 0.020 7,611 0.037 DRAFT – FOR COMMENT, NOT QUOTATION 2.2 Canada VARIABLES Female Born Overseas Speaks other than English Grade level of student Attended Preschool Age started school Parent's level of schooling Mother employed part-time Mother not employed Father employed part-time Father not employed Wealth-standard deviation Constant Observations R-squared (1) Test scores (2) Self-control (3) Perseverance 0.03*** (0.009) 0.01 (0.022) 0.00 (0.018) 0.41*** (0.012) 0.17*** (0.014) 0.05*** (0.007) 0.06*** (0.002) 0.09*** (0.014) -0.01 (0.011) -0.08*** (0.023) -0.02 (0.022) -0.04*** (0.008) -4.79*** (0.117) 0.06*** (0.014) 0.12*** (0.027) -0.03 (0.021) -0.00 (0.019) 0.09*** (0.029) -0.04*** (0.013) 0.03*** (0.004) 0.02 (0.019) 0.01 (0.019) -0.09*** (0.030) 0.00 (0.027) 0.04*** (0.010) -0.19 (0.232) -0.04** (0.015) 0.13*** (0.020) 0.18*** (0.022) 0.14*** (0.020) 0.07** (0.025) -0.02* (0.011) 0.05*** (0.002) -0.01 (0.021) 0.02 (0.019) -0.04 (0.037) 0.04 (0.027) 0.04*** (0.010) -2.08*** (0.212) 15,976 0.097 10,019 0.010 10,026 0.033 DRAFT – FOR COMMENT, NOT QUOTATION 2.3 Ireland VARIABLES Female Born Overseas Speaks other than English Grade level of student Attended Preschool Age started school Parent's level of schooling Mother employed part-time Mother not employed Father employed part-time Father not employed Wealth-standard deviation Constant Observations R-squared (1) Test scores (2) Self-control (3) Perseverance 0.00 (0.017) 0.20*** (0.022) -0.16*** (0.046) 0.08*** (0.011) 0.04 (0.024) -0.09*** (0.014) 0.08*** (0.003) 0.10*** (0.017) -0.03* (0.015) -0.25*** (0.027) -0.18*** (0.016) 0.02 (0.011) -0.98*** (0.157) 0.06*** (0.019) 0.03 (0.036) -0.02 (0.063) -0.01 (0.015) -0.02 (0.024) -0.03* (0.017) 0.03*** (0.004) -0.00 (0.024) 0.06*** (0.021) -0.06 (0.035) -0.17*** (0.029) 0.02 (0.015) -0.25 (0.197) -0.12*** (0.020) 0.08** (0.034) -0.02 (0.071) 0.04*** (0.014) -0.08** (0.031) 0.01 (0.016) 0.04*** (0.004) -0.05* (0.025) 0.01 (0.022) -0.18*** (0.033) -0.04 (0.026) 0.02 (0.014) -0.88*** (0.188) 4,174 0.096 2,741 0.014 2,736 0.020 DRAFT – FOR COMMENT, NOT QUOTATION 2.4 New Zealand VARIABLES Female Born Overseas Speaks other than English Grade level of student Attended Preschool Age started school Parent's level of schooling Mother employed part-time Mother not employed Father employed part-time Father not employed Wealth-standard deviation Constant Observations R-squared (1) Test scores (2) Self-control (3) Perseverance -0.03 (0.021) 0.20*** (0.024) -0.39*** (0.035) 0.41*** (0.024) 0.32*** (0.046) -0.11*** (0.025) 0.11*** (0.005) 0.14*** (0.019) 0.00 (0.021) -0.13*** (0.034) -0.18*** (0.029) 0.08*** (0.013) -5.15*** (0.338) 0.06** (0.022) 0.08*** (0.028) -0.03 (0.039) 0.07** (0.031) 0.13*** (0.036) -0.08*** (0.023) 0.03*** (0.006) -0.01 (0.026) -0.06* (0.031) -0.13*** (0.040) -0.10** (0.045) 0.05*** (0.017) -0.83** (0.345) -0.12*** (0.020) 0.20*** (0.025) 0.31*** (0.032) 0.02 (0.032) -0.01 (0.038) -0.06*** (0.020) 0.04*** (0.007) 0.04 (0.026) -0.14*** (0.025) -0.08* (0.042) 0.12*** (0.037) 0.03** (0.017) -0.59 (0.392) 3,407 0.138 2,169 0.017 2,226 0.049 DRAFT – FOR COMMENT, NOT QUOTATION 2.5 Scotland VARIABLES Female Born Overseas Speaks other than English Grade level of student Attended Preschool Age started school Parent's level of schooling Mother employed part-time Mother not employed Father employed part-time Father not employed Wealth-standard deviation Constant Observations R-squared (1) Test scores (2) Self-control (3) Perseverance -0.02 (0.016) 0.27*** (0.034) 0.04 (0.073) 0.27*** (0.021) 0.32*** (0.060) -0.17*** (0.016) 0.08*** (0.005) 0.08*** (0.023) -0.03 (0.020) -0.19*** (0.038) -0.03 (0.025) 0.04*** (0.012) -3.14*** (0.274) -0.01 (0.024) 0.09* (0.050) 0.22** (0.099) 0.16*** (0.040) 0.30** (0.116) -0.16*** (0.022) 0.03*** (0.006) 0.08*** (0.030) -0.02 (0.033) -0.05 (0.057) -0.07* (0.042) 0.14*** (0.016) -1.69*** (0.546) -0.20*** (0.028) 0.17*** (0.039) 0.51*** (0.075) 0.02 (0.038) 0.28*** (0.091) -0.09*** (0.025) 0.03*** (0.006) 0.04 (0.028) 0.07** (0.026) 0.03 (0.048) -0.03 (0.045) 0.12*** (0.022) -0.46 (0.532) 2,462 0.100 1,602 0.039 1,626 0.033 DRAFT – FOR COMMENT, NOT QUOTATION 2.6 United Kingdom (excluding Scotland) VARIABLES Female Born Overseas Speaks other than English Grade level of student Attended Preschool Age started school Parent's level of schooling Mother employed part-time Mother not employed Father employed part-time Father not employed Wealth-standard deviation Constant Observations R-squared (1) Test scores (2) Self-control (3) Perseverance -0.01 (0.026) 0.19*** (0.036) 0.00 (0.041) -0.02 (0.029) 0.51*** (0.034) 0.06** (0.022) 0.09*** (0.005) 0.17*** (0.017) 0.02 (0.022) -0.11*** (0.031) -0.21*** (0.020) -0.00 (0.011) -1.37*** (0.391) 0.02 (0.021) -0.02 (0.045) 0.17*** (0.048) 0.01 (0.027) 0.11** (0.045) 0.04* (0.018) 0.04*** (0.005) -0.05 (0.030) -0.05 (0.032) -0.03 (0.045) -0.03 (0.038) 0.00 (0.014) -0.85** (0.338) -0.21*** (0.019) 0.29*** (0.043) 0.21*** (0.038) -0.00 (0.023) 0.26*** (0.042) 0.01 (0.018) 0.01** (0.007) 0.07*** (0.026) 0.09*** (0.026) -0.08** (0.037) 0.05 (0.034) 0.06*** (0.011) -0.52* (0.286) 7,263 0.067 4,676 0.009 4,785 0.029 DRAFT – FOR COMMENT, NOT QUOTATION 2.7 USA VARIABLES Female Born Overseas Speaks other than English Grade level of student Attended Preschool Age started school Parent's level of schooling Mother employed part-time Mother not employed Father employed part-time Father not employed Wealth-standard deviation Constant Observations R-squared (1) Test scores (2) Self-control (3) Perseverance 0.02* (0.013) 0.10*** (0.030) -0.18*** (0.019) 0.39*** (0.016) 0.04 (0.052) 0.02** (0.010) 0.06*** (0.003) 0.05** (0.018) 0.04** (0.018) -0.20*** (0.023) -0.14*** (0.020) 0.13*** (0.011) -4.78*** (0.166) 0.14*** (0.016) 0.04 (0.042) -0.05* (0.027) 0.07*** (0.019) 0.30** (0.137) 0.00 (0.014) 0.02*** (0.004) -0.02 (0.039) -0.06** (0.024) -0.12*** (0.035) -0.11*** (0.031) 0.07*** (0.016) -1.24*** (0.258) -0.02 (0.017) 0.11*** (0.039) 0.17*** (0.035) 0.06*** (0.015) -0.17** (0.083) -0.03** (0.016) 0.03*** (0.005) 0.01 (0.031) -0.00 (0.019) -0.17*** (0.043) -0.01 (0.027) 0.10*** (0.012) -0.53** (0.217) 8,627 0.156 5,567 0.023 5,555 0.025 DRAFT – FOR COMMENT, NOT QUOTATION .05 0 -.05 0 is si m no nm is si ng ng is si m ng Pe rs ev er an ce se lf co nt ro l l f-c on tro se l ng is si N on -m Pe rs ev er an ce -.1 -.05 -.1 mean value test scores index .05 Appendix 3: Mean value of test scores by missing value status in non-cognitive abilities for English Speaking countries .4 .3 .2 0 .1 -.3 -.2 -.1 Associated change in index values .5 Appendix 4 unconditional comparisons across English speaking countries Australia Canada Ireland New Zealand Test scores Perseverance UK Scotland Self-Control USA
© Copyright 2026 Paperzz