The Scottish Government

« Previous | Contents | Next »

Listen

SEED Sponsored Research: Children starting school in Scotland

6 Is there any evidence of an optimal age of starting school in Scotland?

6.1 The Dataset

A unique dataset exists for 1289 pupils who were assessed on entry to school and then again three years later in P3. The assessment in P3 was of mathematics, reading, non-verbal ability, vocabulary and of attitudes to maths, reading and school. These data were used to explore the possibility that there might be an optimum age for starting school.

The age of children who took the PIPS BLA and the PIPS P3 assessments had the profile shown below:

Figure 17 Distribution of ages in P3

To a first approximation the distribution is rectangular, corresponding to a single cohort of children with ages 6 months either side of the mean. But there are clearly some exceptions to this general pattern in that there are some pupils who are older than might be expected in the group and a smaller number who are younger. These pupils were presumably delayed for some reason in their entrance to school or moved ahead. Whatever the causes, the pattern is not unexpected and has been seen in other data sets in other countries.

This report concentrates on the cohort of pupils shown in the above distribution and asks questions about their attainment, value-added and attitude scores in the search for evidence of an optimal age for starting school. We further extend this question to include children with different pre-school experience, different home backgrounds and different cognitive profiles.

First the P3 data were explored visually in relation to the age of the children. We then constructed multi-level models to investigate the progress of children between the start of P1 and the end of P3. The models allowed us to look at the link between the age of starting school and the relative achievements in reading and mathematics as well as at the attitudes towards reading, maths and school. Other examples of this approach being employed with PIPS data can be found in (Croxford, 1999, Tymms et al. 2000).

6.1 First look at the data

The P3 measures were plotted against age in the figures below. A line showing the relationship between age and outcome has been added to each plot. This line is a regression line but it is locally weighted ⁷ to show any waves and bumps in the relationships.

Figures 18-21

The charts show little evidence of an optimum age with slight ups and downs for older and younger children. In all charts the highest score is for the children who were five and a half on entry but the link to age is weak. There is also a tendency for the line to fall away for the older children. For non-verbal ability there is a slightly higher score for very young children than for young children.

A more probing question, which asks whether there is an optimum age for value-added, demands additional charts, which are plotted below. The value added measures for the charts were the residuals calculated from a simple regression of maths or reading against the scores on the baseline assessment when starting school.

Figures 22-25

Some waviness is still seen but the rise by age is no longer apparent, although there is still a falling away for the older pupils. The oldest pupils are likely to have repeated an academic year due to low attainment, which is likely to explain the lower residuals for those children.

The visual pattern was then checked using multi-level models.

Table 10 Mathematics and Reading

Table 11 Vocabulary and non-verbal ability

The multilevel models were constructed in such a way that the main predictor of the P3 outcomes (mathematics, reading, vocabulary and non-verbal ability) was the baseline total score, which is the best indicator of later progress. Four dummies were used to indicate (a) the very young children, more than six months below the average age of children starting school, (b) children who were between three and six months younger than the mean starting age, (c) children who were older than the mean starting age by three to six months, and (d) children who were much older. The four separate categories were compared against the children within three months of the average starting age.

None of the coefficients for the dummies in the models were significant and the conclusion therefore is that despite some of the waviness in the lines that we see in the value-added charts above, none of the differences by age were significant. It would appear from this analysis that there is no clear optimum advantage in terms of the progress made by children from their starting point for any particular age on starting school.

6.2 Sex

The data were then checked to see if there was any evidence to suggest that girls or boys were particularly affected by the age of starting school. Terms were introduced in the multi level model for sex and for interactions between sex and the various age categories used above. For each of the four outcomes ten explanatory variables were introduced making 40 in all. Of these, one was significant at the 5% level, indicating that very young girls make less progress than predicted relative to their starting points to the tune of about a fifth of a standard deviation. However, in any large analysis the odd spurious result is to be expected and it is unlikely that this one finding for one subject would be reproduced in further studies.

6.3 Socio-economic status

The home postcodes of the children were linked to the 1991 census data and deprivation indices (Carstairs) were calculated for each neighbourhood from which the pupils originated. The resulting variable was used in the multi level model in two ways. Firstly it was introduced alongside the starting baseline as an additional explanatory variable. This made little difference to the age related coefficients - none were significant. Secondly it was used in combination with each age category to see if children from affluent or deprived backgrounds were particularly advantaged or disadvantaged by starting school at different ages. No evidence for such interactions was found.

6.4 Attitudes

Figures 26 - 27 below show scattergrams of the three measures of attitude in P3 against age. The attitude scales run from 1 to 3 and are formed from an average of the responses children made to a series of statements by selecting frowning (L ), neutral (K ) or smiley (J ) faces. These were coded 1 to 3 respectively.

Visually, there is an indication that the older and younger pupils were slightly more positive for all three outcome measures. Multi-level models were then constructed to quantify the relationships.

Figure 26 Figure 27

Figure 28

Table 11 Attitudes to Mathematics and Reading

None of the coefficients associated with age were significantly different from zero and, in view of the large sample size, it was concluded that the link between age and attitudes seen in the diagrams was very slight and not of educational importance.

6.5 Sex

Girls were generally more positive than the boys. This was particularly true for the Attitude to School measure where the difference was about 0.6 of a standard deviation. However, the multi-level models did not indicate that younger or older boys or girls were particularly positive or negative in their attitudes.

6.6 SES

The multilevel models gave no evidence for deprivation levels changing the conclusions drawn earlier, nor was there any evidence of interactions between age and home background.

6.7 Summary: The age of starting school

No evidence could be found for an optimum age for starting school. More specifically, the cognitive progress and attitudes of children in P3 were unconnected with their age on starting school.

The implications of this are fairly clear so far as changes to policy are concerned: there is no reason to change the age of starting school in Scotland on the basis of the analysis in this report. It provided no evidence that children of four and a half were suffering by starting school too early. Nor did it suggest that five and a half year olds were inappropriately placed. However, this study could not assess the impact of the total amount of schooling on later outcomes at school-leaving age.

« Previous | Contents | Next »