1 of 55

Grace Wallace

M.S. in Psychology

Arizona State University

Advisor: Dr. Nick Duran

Question Based Learning: A Pedagogical Approach for Improving Hypothesis Generation in Science Education

2 of 55

Overview

Background

Benefits of Active Learning
Issues in Hypothesis Generation
Existing pedagogical paradigms

Blended Learning & Guided-Inquiry Learning

Improved Pedagogical Approach: Question-Based Learning
Thesis Study:

Hypotheses, Materials, Methods, Analyses and Results

3 of 55

How Active Learning Works:

Constructive:

Students generate their own understandings

New Information

Prior Knowledge

Content Knowledge

Revise &

Update

Activate &

Evaluate

Hypothesis Generation

Self Regulated Learning

(Azevedo, 2009; Azevedo, 2005; Chi, 2009; Chi & Wiley, 2014)

Active learning tasks are constructive, meaning students are expected to autonomously build, expand, and/or deconstruct mental models of understanding by balancing new information gain with prior knowledge.

To do so requires a multitude of inquiry and metacognitive skills; known as self-regulated learning, these underlying processes are behavioral rather than content based. So, regardless of the academic learning target, the student needs to activate and evaluate their current knowledge base, plan and set goals for their learning, and revise or update existing beliefs throughout the learning task.

These self-regulation skills all contribute to building understanding and content knowledge. In cognitive science, this knowledge-building process is called hypothesis generation. But this process can be susceptible to cognitive biases such as when students are low in prior knowledge, confirmatory analysis, and failure to consider alternatives.

4 of 55

Hypotheses & Misconceptions

Active Learning

Learning Target

Scientific

Misconception

Direct Instruction or

Physical Evidence

(Larsen, Venkadasalam, & Ganea, 2019)

5 of 55

Proactive Approach

Active Learning

Learning Target

Hypothesis Generation

Scientific

Misconception

(Bonawitz et al., 2011; Engel, 2011; Jean et al., 2018; Liquin & Lombrozo, 2017)

6 of 55

How these models fail:

Focusing on a single hypothesis reduces student exploration, critical thinking, and solution evaluation.

Learning Task

Desired Learning Outcome

Unexplored Hypothesis

Presented Hypothesis

Unexplored Hypothesis

(Bonawitz, Shafto, Gweon, Goodman, Spelke, Schulz, 2011; Levy, Thomas, Drago, & Rex, 2013; Markant, 2019)

7 of 55

Improved Approach

Exploratory Questions

Plausible Hypothesis

Desired Learning Outcome

Observe, Experiment, & Revise

(Azevedo, 2009; Azevedo, 2005; Chi, 2009; Chi & Wiley, 2014; Bonawitz et al., 2011; Engel, 2011; Jean et al., 2018; Liquin & Lombrozo, 2017; Roscoe, 2014; Roscoe & Chi, 2007; Ruggeri et al., 2017; Ronfard et al., 2018)

8 of 55

Question-Based Learning (QBL)

Example:

What makes an object fall?
What happens to an object in a gravity-free environment?
How does the momentum of a falling object determine the speed of its fall?
What happens when objects are dropped in a vacuum chamber?

Select question for exploration

Observe, evaluate & revise

jkjlk

Hypothesis Generation

(Azevedo, 2009; Azevedo, 2005; Graesser & McNamara, 2005; Roscoe, 2014; Roscoe & Chi, 2007; Ruggeri et al., 2017; Ronfard et al., 2018)

9 of 55

Question-Based Learning: Evidence

Question-Asking:

Preferred learning mechanism as early as infancy

(Azevedo, 2009; Azevedo, 2005; Bonawitz et al., 2011; Graesser & McNamara, 2005; Lucca & Wilborn, 2019; Roscoe, 2014; Roscoe & Chi, 2007; Ruggeri, Sim, & Xu, 2017 )

Differentiated:

Students select which questions to explore specific to individual need

Manageable:

Children can identify questions best suited for desired information gain

We believe QBL has the potential to be a more effective pedagogical model for a variety of reasons:

First, it relies on a well-known learning mechanism – question-asking – that humans use as early as infancy in the form of pointing gestures to elicit desired information.

Secondly, it also allows individuals to investigate potential hypotheses based on their current predictions and understandings. This necessarily differentiates the hypothesis space on prior knowledge and curiosity, which could encourage individuals to address potential misconceptions during the learning process.

And finally, prior research indicates children can select context relevant and/or highest information gain questions even if they lack the capability to write such questions. This suggests students should be able to select the best questions based on their current understandings; in other words, it helps students manage these self-regulation skills in an open-ended environment.

10 of 55

Thesis Study

Testing our QBL model against two existing forms of pedagogy via a Qualtrics Survey distributed on Amazon’s Mechanical Turk

11 of 55

Thesis Study:

Study 1:

3 randomly assigned pedagogy conditions

Blended Learning
Guided-Inquiry
QBL

All subjects participate in learning task on gas laws
How subjects were guided through task varies based on type of pedagogy model used in each condition

Study 2:

Within Subjects
Same measure
Conditional differences in knowledge decay
Assessment only

Research Question:

Do subjects in QBL condition learn and retain more knowledge than subjects in either the blended learning OR guided-inquiry conditions?

To test the effectiveness of our model design, we conducted a two-part study

During Study 1, subjects were randomly assigned into one of three pedagogy conditions - a blended learning condition, a guided inquiry condition, or a QBL condition - and engaged in a learning task about gas law behavior in kinetic molecular theory. While the measure and learning targets remained the same, how participants were guided through the learning task differed based on the type of pedagogy structure used in each condition. We then conducted a follow-up study with the same subjects, which I will refer to as Study 2, to evaluate conditional differences in knowledge decay. No additional learning task took place in Study 2 - subjects were only asked to retake the assessment measure.

Our research question, therefore, was: “Do subjects in QBL condition learn and retain more knowledge than subjects in either the blended learning OR guided-inquiry conditions?”

12 of 55

Measure

12-question quiz

Learning Targets - Next Generation Science Standards
3 categories of question rigor (low, medium, or high)

4 questions at each level

13 of 55

Low Rigor Question: Example

Which is the best definition of a gas?

Substances that have a fixed amount of stuff that can change shape or size
Substances that have a fixed amount of stuff that cannot change shape or size
Substances that can change shape, size, or amount of stuff inside of them
Substances that are not made of anything and have no stuff inside of them

14 of 55

Medium Rigor Question: Example

If the temperature of a gas decreases, the amount of space that gas takes up…

Remains the same
Decreases
Increases
It depends on the type of gas

15 of 55

High Rigor Question: Example

Say the first image below represents gas particles inside a balloon - Which picture correctly demonstrates the behavior of the gas particles and balloon if temperature is increased?

A.

B.

C.

D.

Original

16 of 55

Measure

12-question quiz

Learning Targets - Next Generation Science Standards
3 categories of question rigor (low, medium, or high)

4 questions at each level

3 time points:

Study 1: Pretest and immediate posttest
Study 2: One-week delayed posttest

17 of 55

Hypotheses

H1: Subjects in the QBL condition will score higher on the post-assessment than those in the blended learning or guided-inquiry conditions.

H2: There will be an interaction effect for condition and assessment rigor such that subjects in the QBL model will demonstrate significantly higher scores on the medium and high rigor questions of the assessment.

H3: There will be an interaction effect for condition and time such that subjects in the QBL condition will retain significantly more knowledge over time as assessed by the week-long delayed measure.

18 of 55

Participants:

Adult sample from MTurk
pwr.t.test function in R (power = 0.80, d = 0.50, a = 0.05)

Paired Samples t-test, n = 27 per condition
Independent-Samples t-test, n = 64 per condition
n = 50 participants per condition

Multiplied by 1.3 to account for participant drop out
Study 1:

n = 65 participants per condition, 195 total
190 subjects completed Study 1

18

19 of 55

Participants:

Study 2: 7-10 days after Study 1

n = 139 returning MTurk participants

Subjects were matched using ID number and birth year

Conditions:

Blended Learning, n = 45
Guided-Inquiry Learning, n = 49
QBL, n = 45

19

20 of 55

Conditions

Blended Learning

Guided-

Inquiry

Question- Based Learning

Suppose you decide to hike to the top of a very tall mountain and pack food for a picnic at the top. When you get there, you notice your bag of chips burst open and there are now chips all over your backpack! You think back to how the same thing happened just a few weeks ago when you left a bag of chips in the car on a hot summer day. You decide to investigate why this keeps happening.

21 of 55

Blended Learning Condition

Direct Instruction

(3 minutes of video)

Real-World

Demo

(2 minutes of video)

8 Follow-Up Questions

What happens when the temperature of a gas is changed?

22 of 55

Guided Inquiry Condition

Real-World Demo

(30-60 sec video)

Answer 2-3 Questions

(9 total)

Repeat

What happens when the pressure of the balloon's environment is changed?

As pressure increases, the balloons get larger
As pressure increases, the balloons get smaller
The balloons remain the same size regardless of pressure

23 of 55

QBL Condition

Exploratory Question

Real-World

Demo

(30-60 sec video)

Follow-Up Question

New Exploratory Question

Posttest

Select One:

24 of 55

QBL Condition

Exploratory Question

Real-World

Demo

Follow-Up Video

How do we know gas particles exist if we can’t see them?

How do we know how gases behave if we can’t see them?

Follow-Up Question

Here is an example for one pathway a subject could have taken in the QBL condition. Let’s say a subject selected the exploratory question “how do we know gas particles exist if we can’t see them?” They would then be shown this real world video demonstration in which an aluminum boat floats on a heavy gas known as sulfur hexafluoride; the idea here is that the subject would note that the aluminum boat has to be held up and - in turn - pushed down by something, meaning that gases particles must exist and contain some kind of mass.

Let’s say our subject still needed further information - on the next screen, they have the option to select this follow-up question “How do we know how gases behave if we can’t see them?” They would then see this corresponding video demonstration in which we see the purple iodine gas in constant motion and expanding to fill the container.

The subject could then return to the original question screen and repeat this process for a different question OR go directly to the posttest.

25 of 55

Condition

Comparison:

Blended Learning:

Uses direct instruction to teach sicentific principle

Guided-Inquiry:

Learn scientific principle by answering given questions

QBL:

Explore selected questions to learn scientific principle

26 of 55

Results & Discussion

Do subjects in QBL condition learn and retain more knowledge than subjects in either the blended learning OR guided-inquiry conditions?

27 of 55

Data Preparation

Data was cleaned and analyzed in R
Question-level analysis for logistic regression

Each observation on the pretest, immediate posttest, and delayed posttest was dummy coded as 0 (incorrect) or 1 (correct)
36 observations per participant

12 at each time point (pretest, posttest, delayed posttest)
5,001 observations total

All R-scripts and data can be accessed via Open Science Framework

28 of 55

Analysis

Bayesian Logistic Mixed Effects Model using brms in R

Dependent Variable (Score)

Predicted probability a subject will answer a given question correctly

Fixed Effects:

Interaction: Condition*Time
Interaction: Condition*Rigor

Random Effects:

Subjects will perform differently at random on pretest, at varying levels of rigor, and across time

Score ~ condition*rigor + condition*time + (1 + rigor + time | subject)

29 of 55

Hypotheses

H1: Subjects in the QBL condition will score higher on the post-assessment than those in the blended learning or guided-inquiry conditions.

Guided Inquiry: b = -.43, OR = .65, 95% CI [-.83 to .05]

Blended Learning: b = .03, OR = 1.03, 95% CI [-.38 to .44]

H2: There will be an interaction effect for condition and assessment rigor such that subjects in the QBL model will demonstrate significantly higher scores on the medium and high rigor questions of the assessment.

Guided Inquiry: b = -.32, OR = .73, 95% CI [-.79 to .17]

Blended Learning: b = .21, OR = 1.23, 95% CI [-.28 to .70]

H3: There will be an interaction effect for condition and time such that subjects in the QBL condition will retain significantly more knowledge over time as assessed by the week-long delayed measure.

Guided Inquiry: b = -.62, OR = .54, 95% CI [-1.03 to -.22]

Blended Learning: b = .19, OR = 1.21, 95% CI [-.23 to .61]

30 of 55

Interaction Effect of Condition*Time with geom_point Overlay

Note: A line graph demonstrating the predicted probability that a subject in each specified condition (blended, guided-inquiry, or QBL) will answer a question accurately at each time point (pretest, posttest, and delayed posttest). Additional data points represent the raw data for individual mean subject scores on the assessment at each time point. Error bars represent 95% confidence intervals.

In hypotheses 1 & 3, we predicted there would be interaction effect of condition and time such that subjects in our QBL condition would perform better than subjects in the blended learning or guided-inquiry learning conditions on the immediate posttest and that these individuals would also be less susceptible to knowledge decay on our delayed posttest. Here, we present a visual representation of this interaction with a geom_point overlay.

To orient you to the graph, our y-axis represents the posterior probability predicted by the model that a subject will answer a given question accurately. On the x-axis we have our time points of pretest, posttest, and delayed posttest. Condition will be indicated by color as depicted in our figure legend, with our QBL condition labeled in red. All error bars represent 95% confidence intervals. We furthermore include a geom_point overlay to represent raw data for individual subject mean assessment scores at each time point.

We can gather several points of information from this graph.

First, there do not appear to be conditional differences in our posteriors at time pretest, indicating random assignment ensured our average participant has similar levels of content knowledge before the learning task regardless of condition.

Furthermore, we see there is a significant increase in our posteriors from time pretest to time posttest. This positive slope indicates that subjects - regardless of condition - should see improved performance on our measure after completing the learning task - providing some evidence that our learning tasks were effective for increasing content knowledge.

31 of 55

Interaction Effect of Condition*Time with geom_point Overlay

Note: A line graph demonstrating the predicted probability that a subject in each specified condition (blended, guided-inquiry, or QBL) will answer a question accurately at each time point (pretest, posttest, and delayed posttest). Additional data points represent the raw data for individual mean subject scores on the assessment at each time point. Error bars represent 95% confidence intervals.

32 of 55

Interaction Effect of Condition*Time with geom_point Overlay

Note: A line graph demonstrating the predicted probability that a subject in each specified condition (blended, guided-inquiry, or QBL) will answer a question accurately at each time point (pretest, posttest, and delayed posttest). Additional data points represent the raw data for individual mean subject scores on the assessment at each time point. Error bars represent 95% confidence intervals.

33 of 55

Interaction Effect of Condition*Time with geom_point Overlay

Note: A line graph demonstrating the predicted probability that a subject in each specified condition (blended, guided-inquiry, or QBL) will answer a question accurately at each time point (pretest, posttest, and delayed posttest). Additional data points represent the raw data for individual mean subject scores on the assessment at each time point. Error bars represent 95% confidence intervals.

34 of 55

Logistic Mixed Effects Model

Note: A triptych interaction plot demonstrating the predicted probability a subject in each specified condition (blended, guided-inquiry, or QBL) at a given time point (pretest, posttest, or delayed posttest) will answer a question accurately for each discrete level of question rigor (low, medium, high). Error bars represent 95% confidence intervals.

In hypothesis 2, we also made predictions regarding the interactions of condition and rigor such that we hypothesized subjects in the QBL condition would be more likely to answer medium and high rigor questions correctly than subjects in the blended or guided-inquiry learning conditions.

Here, we visualize the conditional effects of our logistic mixed effects model, allowing us to see our results at a nuanced level. On the y-axis, we have the predicted probability that a subject will answer a given question accurately and on the x-axis we have our time points of pretest, posttest, and delayed posttest. Condition will be indicated by color as depicted in our figure legend, with our QBL condition again labeled in red. The graph is additionally faceted by rigor so each panel represents a different level of question rigor - low, medium, and high. All error bars represent 95% confidence intervals.

35 of 55

36 of 55

In partial support of our hypotheses, conditional differences do emerge on our immediate posttest.

Drawing our attention to the red line, representing our QBL condition, and the blue line, representing our guided-inquiry condition - we see the error bars overlap for both the low and medium rigor questions. As such, the model predicts there will not be a difference in the probability that a subject in our guided-inquiry or QBL conditions answers a low or medium rigor question correctly after the learning task.

For high rigor questions, however, the non-overlapping error bars do indicate that there is a signficant difference in performance such that the model preducts subjects in the QBL condition have a significantly higher probability of answering a high-rigor question correctly than those in the guided-inquiry condition on the immediate posttest.

Furthermore, we see there are additional differences for subject knowledge decay in the QBL condition and the guided inquiry condition. Non-overlapping error bars indicate that subjects in the QBL condition are more likely to answer low, medium, rigor and high rigor questions correctly than subjects in the guided inquiry condition on the delayed measure.

Taken together, these results suggest our QBL condition learning task improves depth of content knowledge as indicated by better performance on on highly rigorous questions. It additionally appears that subjects in the QBL condition retain significantly more content knowledge than those in the guided-inquiry condition.

37 of 55

Now drawing our attention to the red line - our QBL condition - and the green line - our blended learning condition - we see that our hypothesis was not supported as the overlapping error bars indicate there is no significant difference in predicted accuracy on low, medium, or high rigor questions for subjects at the posttest.

Furthermore, subjects in the QBL condition do not have a higher probability of answering a low or high rigor question accurately on the delayed measure than those in the blended learning conditions.

We do see that subjects in the blended learning learning condition are more likely to answer a medium rigor question correctly after a time delay, which was contrary to our hypothesis. While we did not investigate this difference further, it could be that because subjects in the blended learning condition were both directly told and saw video evidence of gas behavior, it was then easier to recall at a later point and time. It could also be due to the less controlled QBL learning task potentally led subjects to view lots of different video demonstrations and, as a result, it was more difficult to recall specifics regarding gas behavior. Further work is needed to fully investigate this discrepancy

Taken together, these results still indicate there is reason to believe that our QBL approach might be at least as effective as blended learning at teaching content, regardless of rigor and that content knowledge is equally retained over time. .

Our hypotheses were not supported, however, for the blended learning and QBL conditions. Using error bars red line, representing our QBL condition, and the green line, representing our blended learning condition on the high rigor question panel, we see there is no condition-based difference in the probability that subjects will answer a highly rigorous question correctly on the posttest.

Additionally, there is no predicted significant difference in knowledge decay for subjects in the QBL or blended learning conditions for either low or high rigor questions at the delayed posttest. There does appear to be, however, a significant difference on the delayed assessment such that subjects in the blended learning condition have a higher probability of answering a medium rigor question correctly.

While we did not investigate this difference further, it could be that because subjects in the blended learning condition were both directly told how gas behavior changes and saw video evidence of this behavior, it was then easier to recall at a later point and time. It could also be that because of being engaged in a less controlled learning task, subjects in the QBL condition potentially saw lots of different video demonstrations and, as a result, it was more difficult to recall specifics regarding gas behavior. Further work is needed to fully investigate this discrepancy

Taken together, these results still indicate there is reason to believe that our QBL approach might be at least as effective as blended learning at teaching content, regardless of rigor and that content knowledge is equally retained over time. .

38 of 55

Discussion:

The QBL model appears to be as effective as blended learning models at increasing assessment scores at all levels of rigor and this learning is retained over time

The QBL model appears to be more effective than guided-inquiry learning at increasing performance on high level rigor questions and more knowledge gained from the learning task is maintained over time

39 of 55

Limitations:

Use of online platform with adult subjects may not generalizable to students in classrooms
Validity of data measure

Hypothesis generation might be better suited to qualitative analysis

Implementation Feasibility

QBL models provide significantly less control, which could reduce learning efficiency or be limiting to students unable to engage in this process

We do recognize some limitations in our work.

First - the use of an online platform with adult learners may not be generalizable to students in classrooms. We recognize that while our work is a starting point, significantly further research needs to be done to determine whether this model would be appropriate for school-aged children.

Furthermore, using a multiple choice assessment might lack validity as a measure for hypothesis generation. While other work in hypothesis generation utilizes similar measures, it is possible that a qualitative analysis would be better suited because it would give direct insight into the subject’s line of thinking.

Finally, we do also recognize the implementation feasibility for our QBL model in classrooms. The model’s open-ended design might not be the best approach for all content or students; rather, we suggest our QBL model be approached as an additional tool in the pedagogical toolbox.

40 of 55

Contributions:

Demonstrates the effectiveness of student directed inquiry and exploration for building content knowledge

Brings a classroom-based pedagogical approach for using questioning as a learning mechanism

Improving upon past work with teacher questioning, tutor-peer interactions, and computer-based learning

Furthers past research on the need to scaffold self-regulated learning in open-ended exploration

(Azevedo, 2005; Azevedo, 2009; Chi, 2009; Chi & Wily, 2014; Grasser et. al, 2005; Levy et al., 2013; Markant et al., 2016, Markant 2018)

41 of 55

Conclusions:

QBL model appears to be an effective form of active learning pedagogy
Future Work:

Diversified scientific principles
Qualitative analysis of generated hypotheses
Replicate the results in an offline distribution platform and with school-aged children.

42 of 55

THANK YOU!

43 of 55

Azevedo, R. (2005). Using hypermedia as a metacognitive tool for enhancing student learning? The role of self-regulated learning. Educational psychologist, 40(4), 199-209.

Azevedo, R. (2009). Theoretical, conceptual, methodological, and instructional issues in research on metacognition and self-regulated learning: A discussion. Metacognition and Learning, 4(1), 87-95.

Berlyne, D. E. (1954). A theory of human curiosity. British Journal of Psychology, 45, 180–191.

Bonawitz, E., Shafto, P., Gweon, H., Goodman, N. D., Spelke, E., & Schulz, L. (2011). The double-edged sword of pedagogy: Instruction limits spontaneous exploration and discovery. Cognition.

Chi, M. T. (2009). Active-constructive-interactive: A conceptual framework for differentiating learning activities. Topics in Cognitive Science, 1(1), 73-105.

Chi, M. T., & Wylie, R. (2014). The ICAP framework: Linking cognitive engagement to active learning outcomes. Educational Psychologist, 49(4), 219-243.

Engel, S. (2011). Children's need to know: Curiosity in schools. Harvard educational review.

Fandakova, Y., Lindenberger, U., & Shing, Y. L. (2015). Episodic Memory Across the Lifespan. The Wiley handbook on the cognitive neuroscience of memory, 309.

Freeman, S., Eddy, S. L., McDonough, M., Smith, M. K., Okoroafor, N., Jordt, H., & Wenderoth, M. P. (2014). Active learning increases student performance in science, engineering, and mathematics. Proceedings of the National Academy of Sciences, 111(23), 8410-8415.

Fox, A. R., Hollan, J. D., & Walker, C. M. (2019) When graph comprehension is an insight problem. In A. Goel, C. Seifert, & C. Freska (Eds.), Proceedings of the 41st Annual Meeting of the Cognitive Science Society. Montreal, CA: Cognitive Science Society.

Graesser, A. C., McNamara, D. S., & VanLehn, K. (2005). Scaffolding deep comprehension strategies through Point&Query, AutoTutor, and iSTART. Educational psychologist, 40(4), 225-234.

Jauhariyah, M. N. R., Suprapto, N., Admoko, S., Setyarsih, W., Harizah, Z., & Zulfa, I. (2018, March). The Students’ misconceptions profile on chapter gas kinetic theory. In Journal of Physics: Conference Series (Vol. 997, No. 1, p. 012031). IOP Publishing.

References

44 of 55

Jean, A., Daubert, E., Yu, Y., Shafto, P., & Bonawitz, E. (2019, January). Pedagogical questions empower exploration. In Proceedings of the Annual Conference of the Cognitive Science Society.

Kang, M. J., Hsu, M., Krajbich, I. M., Loewenstein, G., McClure, S. M., Wang, J. T. Y., & Camerer, C. F. (2009). The wick in the candle of learning: Epistemic curiosity activates reward circuitry and enhances memory. Psychological Science.

Larsen, N.E., Venkadasalam, V.P., & Ganea, P.A. (2019). Without conceptual information children miss the boat: Examining the role of explanations and anomalous evidence in scientific belief revision. In A. Goel, C. Seifert, & C. Freska (Eds.), Proceedings of the 41st Annual Meeting of the Cognitive Science Society. Montreal, CA: Cognitive Science Society.

Levy, B. L., Thomas, E. E., Drago, K., & Rex, L. A. (2013). Examining studies of inquiry-based learning in three fields of education: Sparking generative conversation. Journal of Teacher Education, 64(5), 387-408.

Liquin, E., & Lombrozo, T. (2017). Explain, Explore, Exploit: Effects of Explanation on Information Search. CogSci.

Litman, J. (2005). Curiosity and the pleasures of learning: Wanting and liking new information. Cognition & emotion, 19(6), 793-814.

Loewenstein, G. (1994). The psychology of curiosity: A review and reinterpretation. Psychological Bulletin, 116(1), 75–98.

Lucca, K. & Wilbourn, M.P. (2018). The what and the how: Information-seeking gestures facilitate learning object labels and functions. Journal of Experimental Child Psychology.

Markant, D. B., Ruggeri, A., Gureckis, T. M., & Xu, F. (2016). Enhanced memory as a common effect of active learning. Mind, Brain, and Education, 10(3), 142-152.

Markant, D. B. (2018). Effects of biased hypothesis generation on self-directed category learning. Journal of Experimental Psychology: Learning, Memory, and Cognition.

Nakhleh, M. B. (1992). Why some students don't learn chemistry: Chemical misconceptions. Journal of chemical education, 69(3), 191.

PCAST STEM Undergradaute Working Group (2012) Engage to Excel; Producing One Million Additional College Graduates with Degrees in Science, Technology, Engineering, and Mathematics, eds Gates SJ, Jr, Handelsman J, Lepage GP, Mirkin C (Office of the President, Washington).

References

45 of 55

Pluck, G., & Johnson, H. L. (2011). Stimulating curiosity to enhance learning. GESJ: Education Sciences and Psychology, 2.

Preston, A. R., & Eichenbaum, H. (2013). Interplay of hippocampus and prefrontal cortex in memory. Current Biology, 23(17), R764-R773.

Ronfard, S., Zambrana, I. M., Hermansen, T. K., & Kelemen, D. (2018). Question-asking in childhood: A review of the literature and a framework for understanding its development. Developmental Review.

Roscoe, Rod & Chi, Michelene. (2007). Understanding Tutor Learning: Knowledge-Building and Knowledge-Telling in Peer Tutors' Explanations and Questions. Review of Educational Research - REV EDUC RES. 77. 534-574.

Roscoe, R. D. (2014). Self-monitoring and knowledge-building in learning by teaching. Instructional Science, 42(3), 327-351.

Ruggeri, A., Markant, D. B., Gureckis, T. M., Bretzke, M., & Xu, F. (2019). Memory enhancements from active control of learning emerge across development. Cognition, 186, 82-94.

Ruggeri, A., Sim, Z. L., & Xu, F. (2017). “Why is Toma late to school again?” Preschoolers identify the most informative questions. Developmental psychology.

Shah, P. E., Weeks, H. M., Richards, B., & Kaciroti, N. (2018). Early childhood curiosity and kindergarten reading and math academic achievement. Pediatric research.

Shing, Y. L., & Brod, G. (2016). Effects of prior knowledge on memory: Implications for education. Mind, Brain, and Education, 10(3), 153-161.

Stahl, A. E., & Feigenson, L. (2015). Observing the unexpected enhances infants’ learning and exploration. Science.

Webb, N. L. (1997). Criteria for Alignment of Expectations and Assessments in Mathematics and Science Education. Research Monograph No. 6.

References

46 of 55

H1: Subjects in the QBL condition will score higher on the post-assessment than those in the blended learning or guided-inquiry conditions.

Guided Inquiry:

b = -.43, OR = .65, 95% CI [-.83 to .05]

Blended Learning:

b = .03, OR = 1.03,

95% CI [-.38 to .44]

47 of 55

H2: There will be an interaction effect for condition and assessment rigor such that subjects in the QBL model will demonstrate significantly higher scores on the medium and high rigor questions of the assessment.

Guided Inquiry:

b = -.32, OR = .73, 95% CI [-.79 to .17]

Blended Learning:

b = .21, OR = 1.23, 95% CI [-.28 to .70]

48 of 55

H3: There will be an interaction effect for condition and time such that subjects in the QBL condition will retain significantly more knowledge over time as assessed by the week-long delayed measure.

Guided Inquiry:

b = -.62, OR = .54, 95% CI [-1.03 to -.22]

Blended Learning:

b = .19, OR = 1.21, 95% CI [-.23 to .61]

49 of 55

Grace Wallace

M.S. in Psychology

Arizona State University

Advisor: Dr. Nick Duran

Question Based Learning: A Pedagogical Approach for Improving Hypothesis Generation in Science Education

50 of 55

Overview

Background

Benefits of Active Learning
Issues in Hypothesis Generation
Existing pedagogical paradigms

Blended Learning & Guided-Inquiry Learning

Improved Pedagogical Approach: Question-Based Learning
Thesis Study:

Hypotheses, Materials, Methods, Analyses and Results

51 of 55

Passive vs. Active Learning

Passive Learning:

Lecture Style
Teacher-directed

What, when, and how information is transmitted

Driven by practice, memorization, and recall
Students are “passive”

Active Learning:

Open-ended Style
Student-directed

Discover learning targets through exploration & inquiry

Driven by curiosity, reasoning, and revision
Students are “active”

Cognitive Benefits:

Memory retention
Deeper conceptual knowledge

Academic Benefits:

Increased GPA
Lowered failure rates

(Chi, 2009; Chi & Wiley, 2014, Levy et al., 2013; Freeman et al., 2014 ; Markant, 2018)

In the past, passive learning lectures that gave teachers authority over what, when, and how new information was transmitted to students. In such environments, knowledge gain was driven by drilled practice, memorization, and factual recall.

Contemporary K-12 classrooms, however, focus on increasing student engagement and applied skills with an open-ended style of lesson design known as active learning. These lessons are student-directed, asking students to autonomously self-discover learning targets through exploration and inquiry. Here, learning is driven by increased student curiosity, questioning, and reasoning.

Past research demonstrates this shift promotes both cognitive benefits including increased memory retention and deeper conceptual understandings as well as improved academic achievement in science-related courses.

52 of 55

How Active Learning Works

Collaboration:

Opportunities for feedback and reflection

Constructive:

Students generate their own understandings

(Chi, 2009; Chi & Wiley, 2014)

Pace:

Gives students control over input and processing

53 of 55

Hypothesis Generation

Learning Target:

Gravitational acceleration of any free-falling object is at a constant 9.8m/s²regardless of object’s size

1-lb

10-lb

(Levy, Thomas, Drago, & Rex, 2013; Markant, 2019; Markant et al., 2016)

Low Prior Knowledge: Objects with higher momentum fall with greater acceleration

Confirmatory Analyses: Fail to account for air resistance when lighter objects fall slower

Selective Memory Retrieval

Failure to Consider Alternatives

In a perfect world, learners would have these underlying processes and be able to reach accurate content understandings simply by engaging in any active learning task. The reality, however, is that hypothesis generation is much more nuanced, especially in science education.

Let’s take an example using an essential learning target in introductory physics.

Teachers frequently start this lesson with the question, “If a heavy object like a bowling ball and light object like a feather are dropped at the same time in a world without air resistance, which hits the ground first and why?”

Most students predict the heavier object will fall faster and, therefore, hits the ground first. This belief, however, represents a common misconception – the two objects actually fall with the same acceleration and would hit the ground at the same time.

Misconceptions such as these are rooted in several cognitive biases that can occur during the process of hypothesis generation. For example, students low in prior knowledge might assume the object falling with a higher momentum will automatically fall with a greater acceleration. Other students might, instead, fall victim to confirmatory analyses; I know if I drop a basketball and posterboard, for example, the basketball falls faster and therefore the bowling ball would fall faster in this example. Other biases in hypothesis generation include selective memory retrieval and failure to consider alternatives.

54 of 55

Current Models

Blended Learning

Direct instruction followed by active learning
“Cookbook” Teaching – passive during instruction
High efficiency
Low constructivist portion of active learning

(Levy, Thomas, Drago, & Rex, 2013)

“All objects fall at 9.8m/s^2”

Practice Dropping Objects

Prove given hypothesis

Today, we are going to examine two existing types of pedagogy commonly implemented by K-12 science teachers to avoid issues in hypothesis generation. At this point, I will just outline the basic paradigm of how these models work, however, an example will be provided when discussing conditional differences in the final thesis study.

Blended learning is a “cookbook” style of teaching, because students learn through direct instruction and later actively engage with via experimentation - similar to cooking where the chef knows the desired outcome and then follows directions to produce that desired outcome.

Although providing students the learning target directly largely removes the potential for misconceptions and is highly efficient for disseminating information, it also significantly reduces the amount of student-direction in that constructivist portion of active learning.

Blended learning is where students are directly taught the learning target by an instructor and then engage in an active learning task to prove said principle. Such methodology is often referred to as “cookbook” teaching because students passively learn information that they later actively engage with via experimentation - similar to cooking where the chef knows the desired outcome and then follows directions to produce that desired outcome.

55 of 55

Current Models

Guided-Inquiry Learning

Scaffolded questioning guides learner through task
Learn by answering teacher’s questions
Moderate efficiency
Moderate constructivist portion of active learning

(Levy, Thomas, Drago, & Rex, 2013)

Practice dropping objects

What does it mean that both objects hit the ground at the same time?

Prove given hypothesis