Issues of Causality in Childrens Comprehension Development

1
Issues of Causality in Childrens Comprehension
Development

• Jane Oakhill (University of Sussex)
• Kate Cain (University of Essex)

2

• How do we know which skills and strategies to
  train?
• We need to find out which associates of
  comprehension skill are likely to be causally
  implicated in its development.
• How can we find out?
• (A. by doing training studies!)

3
Ways to explore causal hypotheses

• The Comprehension-Age Match design a design
  analogous to the Reading-Age Match design.
• Longitudinal studies.
• Training studies.

4
The Comprehension-Age Match design

• Analogous to the Reading Age Match design
  pioneered by, e.g. Goswami and Bryant.
• If the CAM group are better at (for example)
  inference making than older poor comprehenders,
  then we can rule out the possibility that
  superior inference skill results from superior
  comprehension because the CAM group and
  less-skilled comprehenders are matched on
  absolute level of comprehension skill.

5
Characteristics of subjects in CAM design
6
Characteristics of subjects in CAM design
7
Inference skill and comprehension ability

• Good and poor comprehenders differ in their
  propensity to make (at least) two different types
  of inference
• Integration of information provided in the text,
  to establish coherence (text-connecting
  inferences)
• Incorporation of information from outside the
  text (general knowledge) with information in the
  text to fill in missing details and understand
  the text as a whole (gap-filling inferences)

8
Example study using the CAM Inferences

• Debbie was going out for the afternoon with her
  friend Michael. By the time they got there they
  were very thirsty. Michael got some drink out of
  his duffel bag and they shared that. The orange
  juice was very refreshing. Debbie put on her
  swimming costume but the water was too cold to
  paddle in, so they made sandcastles instead.
• They played all afternoon and didn't notice how
  late it was. Then Debbie spotted the clock on
  the pier. If she was late for dinner her parents
  would be angry. They quickly packed up their
  things. Debbie changed and wrapped her swimming
  costume in her towel. She put the bundle in her
  rucksack. Then they set off for home, pedaling as
  fast as they could. Debbie was very tired when
  she got home, but she was just in time for
  dinner.

9
Example questions

• Literal information
• Who did Debbie spend the afternoon with?
• Where was the clock?
• Text-connecting (bridging) inferences
• Where did Michael get the orange juice from?
• Where did Debbie put her towel when she packed up
  her things?
• Gap-filling inferences
• Where did Debbie and Michael spend the afternoon?
• How did Debbie and Michael travel home?

10
Inference study Percentage of correct responses
for each skill group.
11
Inference study Percentage of correct responses
for each skill group.
12
CAM inference study Summary and conclusions.

• Less-skilled comprehenders were poorer at
  answering inferential questions than skilled
  comprehenders, and were worse than the CAM group
  at text-connecting inferences.
• Therefore, skill at drawing inferences is not
  simply a by-product of reading comprehension
  skill it is a candidate cause of good
  comprehension.

13
Story production task

• Picture prompt condition
• Child given a series of pictures on which to base
  a story.
• Topic prompt condition
• Child given a topic, e.g. Pirates, The
  holiday on which to base a story.

14
Results

• The productions were classified using a three-way
  classification which reflected level of
  interconnectedness and causality. The categories
  were
• non-story
• intermediate
• well-formed stories
• The childrens productions were given a score
  according to which category they fell into, and
  these scores were used as DVs in the analysis.

15
Examples of story productions (all from topic
prompt)

• Non story (the holiday)
• Once upon a time there was a girl and she went on
  holiday.
• Intermediate story (the seaside)
• It was a lovely day. The family decided to go
  down to the seaside. They saw lots of people
  there. The baby was making a sandcastle. The
  older children were playing in the sea. The mum
  and dad had their last swim before they went home.

16
Examples of story productions (all from topic
prompt)

• Complete story (the circus)
• One day there was a girl named Gigi. She went to
  a circus. She saw loads of things she liked.
  The best thing she liked was the lion... In the
  morning she found the lion she liked the best,
  what was from the circus, walking up and down the
  road by her house. So she said, Mum, the lion I
  saw in the circus last night is walking up and
  down in the road. So her mum said, Go and get
  some kippers and try and head him back to the
  circus and put him in the cage. So she did. The
  end.

17
Story production average number of words
18
Example study using the CAM Story structure

• In this study, children were asked to produce
  stories. The study had various aims, but the
  present interest is in differences in the quality
  of the story structures between the groups.
• The groups were similar to those in the inference
  study.

19
Story production scores (max6)
20
CAM story production study summary and
conclusions

• Less-skilled comprehenders produced less
  well-structured stories than did skilled
  comprehenders, and were worse than the CAM group
  in a condition where only a topic for the story
  (e.g. Pirates) was provided .
• Therefore, the ability to produce causally
  coherent stories is not simply a by-product of
  reading comprehension skill it is more likely to
  reflect a cause.

21
CAM Design Conclusions, assumptions limitations

• The CAM design can be used to test the
  plausibility of causal hypotheses, and to rule
  out some possible hypotheses.
• The CAM design can only be used to rule out
  possible causal links, not to prove them.
• Positive findings can then be followed up with
  more costly and time-consuming designs, such as
  longitudinal and/or intervention studies.

22
Longitudinal prediction of reading skills

• Children were assessed on a variety of skills and
  abilities at three time points
• Year 3 (7-8 years)
• Year 4 (8-9 years)
• Year 6 (10-11 years)
• We can explore the relation between early skills
  and later reading ability.

23
Longitudinal study questions

• Do different sub-skills predict comprehension and
  word reading accuracy across time? (yes)
• Do different sub-skills of comprehension
  independently account for variance in later
  comprehension skill?

24
Longitudinal studySkills and abilities measured

• DVs Neale Comprehension, Neale Accuracy
• Verbal and performance IQ (T1T3 only)
• Vocabulary (BPVS)
• Syntax (TROG)
• Phonological skills phoneme deletion and oddity
• Working memory sentence span and digit task
• Inference skill and literal memory for text
• Comprehension monitoring
• Story structure understanding (anagram task,
  titles task, main point identification).

25
Path diagram to show the significant Time 1 and
Time 2 predictors of comprehension at Time 3 .
ComprT2
ComprT1
ComprT3
VIQ
BPVS1
Inference2

Story structure1


Compr. Monitor2
Compr. Monitor1
26
Conclusions from longitudinal study

• Specific sub-skills of comprehension inference
  skill, story structure understanding,
  comprehension monitoring predict later reading
  comprehension (but not word reading).
• Each of these sub-skills is predictive of later
  comprehension skill (but not word reading
  accuracy) over and above the autoregressive
  effect of comprehension, and in competition with
  general ability measures (VIQ, vocabulary,
  working memory).

27
Conclusions Implications

• The study confirms that a set of higher-level
  comprehension components, which, on theoretical
  grounds, ought to be instrumental in the growth
  of reading comprehension skill, may indeed be
  instrumental.
• These findings also suggest that comprehension
  does not necessarily develop automatically once
  word decoding is proficient, but that it is
  dependent on different skills, and may need
  specific teaching.

28
Within-population variation

• Recent studies have shown that some poor
  comprehenders can have relative strengths in key
  comprehension-related skills (as well as
  weaknesses).
• E.g. Cornoldi et al. (1996) found that not all
  poor comprehenders were poor at comprehension
  monitoring.
• Cain Oakhill (in press) found substantial
  heterogeneity in a population of poor
  comprehenders.

29
Training study

• Our training studies have focused on processing
  strategies and, in particular,ways to think about
  the text if it relates to what one knows,
  whether understanding is adequate.
• General idea was to get children more aware of
  and more involved in their own comprehension --
  to encourage inferences and monitoring.
• Based on Palincsar and Browns reciprocal
  teaching.
• Training resulted in improvements (on
  standardized test) for poor (not good)
  comprehenders, compared to control training.

30
Components of training

• Trained group received training in 7x30min.
  Sessions in small sub-groups.
• Inferences from single words in sentence (all 7
  sessions)(e.g. the setting, or something about a
  character)
• Question generation - children took turns, and
  discussed answers. (4 sessions)
• Prediction - what might come next? Text revealed
  and predictions discussed (1 session).
• Two control groups comprehension exercises
  (again in small groups, with discussion of
  responses) and automated word decoding.

31
Results of training study

• Short term training increased performance on
  standardized reading comprehension test.
• The less-skilled comprehenders improved more than
  the good comprehenders either with
  inference/question generation training or
  comprehension exercises.
• The less-skilled comprehenders given inference
  training improved more than those given decoding
  training.

32
Results of training study mean comprehension
improvement in months
33
Conclusions

• So - what ARE these skills that underlie
  comprehension, and that might be amenable to
  training?
• Three we have identified are
• Inference making
• Comprehension monitoring
• Understanding story structure
• Could these be part of some more general ability?
• Should they be trained independently or together?

34
Conclusions (contd)

• A relation between inference making,
  comprehension monitoring, story production and
  comprehension could be partly mediated by the
  readers standard for coherence.
• For comprehension to develop to higher levels,
  the reader must adopt a high standard of
  coherence to care whether the text makes sense.
  
• In other words, readers must strive to derive a
  clear, complete and coherent situation model of
  the text.

35
Conclusions (contd)

• When coherence is a goal, inferences are made to
  keep the text coherent.
• When coherence is a goal, inconsistencies between
  text elements or between text elements and the
  readers knowledge are resolved rather than
  ignored or not noticed.
• When coherence is a goal, the point and structure
  of the text as a whole will be appreciated by the
  reader, and will in turn, guide and reinforce
  their comprehension.

36
Conclusions (contd)

• So, should a goal of strategy training be to
  encourage children to adopt a high standard for
  coherence when they are reading as the default?
• And, if so, how can this training best be
  achieved?

37
Consistent with NICHD reports conclusions

• The NICHD report (2000) identified seven
  categories of comprehension instruction that have
  solid evidence for their effectiveness.
• These seven include procedures that draw the
  reader into deeper engagement with the text -
  i.e. more active processing.
• They include comprehension monitoring, question
  answering, question generation, use of semantic
  organizers, and student summarization.
• All these are strategies which encourage readers
  to adopt a high standard for coherence.

38

• And this view is consistent with the sort of
  studies that we are going to hear much more about
  today and tomorrow!
• So - we seem to have come full circle - the
  research we are doing on correlates and causes
  converges with the results from instructional
  studies.

39
Some questions

• Can comprehension best be trained by training
  individual comprehension skills ?
• Or is a more integrated training approach better?
• And, do different individuals benefit from
  training in some skills, not others?