Probabilistic
Intuitions in Traditional vs. Alternative ContextS
Dale Havill, Ph.D. Educational Psychology
Zayed University, United Arab Emirates
DaleHavill@bigfoot.com
Research and assessment of
probabilistic reasoning has often presented text-only multiple choice problems
which ask students to reason about common random
generating devices, (e.g., coins, dice). This study presented problems in
three contexts: traditional (“COIN/DIE”), pinball diagrams ("GALTON")
and database frequencies ("PLANET”). Research questions focused on a
common source of students' errors and biases: the distribution of probabilities
associated with ordered sequences of
independent events. Results indicated that general purpose cognitive heuristics
described in previous research change or are not as salient in some alternative
problem contexts. Students' problem solving frameworks, heuristics, and
atmospheric beliefs were interpreted by extending Fischbein's theoretical
framework of primary (everyday) and secondary (instruction-related)
intuitions.
In recent
years educational organizations have given increased attention to probability
curricula and have recommended introduction of probability concepts and
activities at earlier grade levels (NCEE, 1995; NCTM, 2000). However, students'
difficulty in reasoning normatively1 has posed considerable challenges for curriculum
design and implementation (Algren & Garfield, 1991; Jones et al., 1999;
Shaughnessy, 1992). These pedagogical challenges are due in part to the fact
that, unlike many mathematical topics, naive intuitions (i.e., immediate
apprehensions or cognitions2) about
probability are entrenched in everyday discourse and activity. Cognitive as
well as educational researchers have noted that there are large gaps in our
understanding of how these intuitions interact with instruction-based
knowledge, and how best to deal with intuitions in order to facilitate
students' construction of normative concepts and problem solving skills.
Research
Questions
Traditional
cognitive research investigating probabilistic reasoning has often focused on
intuitive errors and biases which indicate that human fallibility is associated
with natural or everyday thinking. However, many traditional research
instruments utilized only limited response sets (e.g., multiple-choice
responses). Furthermore, problem situations presented in traditional studies
typically involved only standard random generating devices (e.g., coins, dice,
spinners, etc.). For example, consider traditional versions of the three problems
presented in this study (correct choices for the three problems are 1-c, 2-e,
and 3-b):
1. Referred to below as the "Four Previous Heads" problem:
If
a coin is flipped four times and comes up heads every time, which is more
likely on the next flip?
a.
head
b.
tail
c.
head and tail equally likely
2. Referred to below as the "Equiprobable Ordered
Sequences" problem:
Which
sequence of coin flips is most likely?
a.
HHHTT
b.
THTTH
c.
HHHTH
d.
HTHTH
e.
All sequences are equally likely.
3. Referred to below as the "Skewed Probability Samples"
problem:
A
die is painted black on five sides and white on one side. What is the most
likely outcome of six rolls?
a.
6 blacks and no whites
b. 5 blacks and
one white
FIGURE 1. Basic
multiple choice versions of the problems presented in this study.
A primary
motivation for the study was to explore how students' responses differ in
alternative contexts. In particular, do naïve intuitive responses found in heuristics and biases research vary
across the three contexts? Fischbein (1975) suggested that investigating such
differences can help us identify secondary
intuitions (i.e., acquired in traditional contexts through school
curricula) as well as primary intuitions
(i.e., general purpose heuristics and biases acquired through everyday
experience). In other words, school-based instruction—as well as general
purpose everyday cognitive processes—can be a significant source of
misappropriated intuitions. Of particular interest are the Equiprobable Ordered Sequences and Skewed Probability Samples problems in which students appear to
misappropriate formal probabilistic knowledge. Intuitions about ordered sequences, (in which
multiplicative-analytical skills facilitate understanding that probability is
related to conditional dependence of events in a sequence), can erroneously
substitute for, or be substituted by, intuitions about unordered samples (in which the probability is directly related to
the 'partitive' proportion of independent outcomes in the sample).
Theoretical
and Historical Framework
The
traditional normative theoretical
framework provided a basic measure for evaluating students’ judgments in this
study—a response was considered correct if it was consistent with generally
accepted mathematical models of probability. However, even when responses are
clearly correct or incorrect in a normative framework, theorists are sometimes
divided into opposing camps depending on whether human thinking is
characterized as more, or less, rational. On one hand, early cognitive
investigations provided evidence which indicated that everyday heuristics and biases are irrational
obstacles to normative probabilistic reasoning (Kahneman et al., 1982). On the
other hand, some researchers contended that a critical analysis of experimental
paradigms is crucial to interpreting research that focused on erroneous
heuristics and biases:
The paradigms [are] so limited and inadequate that
generalisations from current research on heuristics and biases cannot be
justified. In particular, the view of people as 'intellectual cripples', who
exhibit severe and systematic biases in making judgements, [is] a value
judgement on the part of the investigators.
(Phillips, 1983, p. 525)
In
general, theories of probabilistic reasoning have differentiated 1)
frequentist-distributional vs. local-atmospheric thinking, and 2) intuitive vs.
analytical thinking. Interpretation of results in this study was influenced by
Fischbein's (1975) theoretical framework which viewed 'intuitional' cognition
to be as important for human functioning as formal-operational cognition (cf.
the Piagetian focus on formal-operational cognition as the end state in the development of higher order thinking).
Research
Method
Virtually all 168 participating undergraduate
students were drawn from psychology department subject pools in two U.S.
universities. Students were given a question sheet packet and worked alone;
most finished the problems in about ten to fifteen minutes. Each question sheet
packet contained three problems presented in only one context.
Figure 1 shows the basic structure of problems
presented in the traditional "COIN/DIE" context and Figure 2 gives a
basic idea of how diagrams looked in the alternative contexts problem (note
that the diagrams in Figure 2 are small iconic versions of the diagrams
presented to students). Problems in the "GALTON" context showed
diagrams of a simple pinball machine in which the path of a pinball was traced
through a triangular set of pins (a right/left branch of the pinball is
analogous to head/tail in a coin flip). Problems in the "PLANET"
context described space exploration probes whose data indicated that there was
a 50% chance that planets in a solar system were inhabitable (the 50% chance
analogous to head/tail in the traditional context).
In addition to reasoning about the diagrams students
were asked to estimate frequencies and percentages for every multiple-choice
alternative, and to write a short justification of the alternative they
selected. These multiple representations and response modes were designed to
facilitate correct interpretation and evoke a more informative response set.
Two general outcomes related to problem format and context will be briefly described, followed by results showing how responses in the alternative contexts differed from the heuristics and biases model of reasoning in traditional contexts. A first general observation in considering the overall response sets is that problem formats with multiple representations and modes of response seemed to have little positive effect. For example, many students gave frequency and percentage estimations that were incoherent (e.g., estimated percentages weren't consistent with estimated frequencies, or the sum of percentage estimations for each alternative was over 100%, even when alternatives were a subset of all possible outcomes).
Table 1
Percent Correct On Multiple Choice Responses In Each
Context and Problem Type
|
Context |
N |
Four
Previous Heads |
Equiprobable Ordered
Sequences |
Skewed
Probability Samples |
|
COIN/DIE |
56 |
93 % |
27 |
38 |
|
GALTON |
57 |
63 |
32 |
41 |
|
PLANET |
55 |
45 |
2 |
62 |
A second general observation involved differences
across contexts in students' willingness to accept assumptions about the
independence of a future event. Performance on the traditional version of the Four Previous Heads problem (93%
correct) indicated that most students are familiar with and willing to accept
the idea that a future coin flip is independent of previous events. In
contrast, students weren't as likely to accept assumptions about independence
of events in nontraditional contexts (63% correct in the GALTON context, and
45% correct in PLANET context). Students tended to hypothesize physical causes
in the GALTON context and nonprobabilistic dependencies in the PLANET context:
GALTON
student: “The ball has momentum and will
continue going in that direction."
PLANET
student: “Since one planet is
inhabitable the others are more likely to be inhabitable.”
The situated nature of students' reasoning in the
alternative contexts was interesting because the study was designed in part to
explore how computer-based activities in the alternative contexts might help
students reason normatively.
Significant
differences from traditional results may be related to how events are
represented as well as students' tendency to hypothesize nonrandom causes or
associations in the alternative contexts. Figure 2 shows which outcomes
students thought were most likely in the Equiprobable
Ordered Sequences problem. First note that results in the traditional
COIN/DIE context were consistent with the heuristic
and biases model. Outcome #2 (“THHTH”) tends to be viewed as most likely because it is
more 'representative' of a random process: 68% of students in the COIN/DIE
context ranked "THHTH" among the two most probable outcomes. The
other alternatives tend to be viewed as less representative of a random process
because “HHHTT” has runs, “THTTT” has a lopsided proportion of heads, and “HTHTH”
alternates too regularly (see rankings in Figure 2).
Students
in the GALTON and PLANET contexts more often chose Outcome #1 ("HHHTT"
in the traditional context) as most likely. Moreover, the second highest
ranking in both alternative contexts was Outcome #3 ("THTTT" in the traditional context) which has a run as well
as the least balanced proportion. Thus in contrast to results found in
traditional studies, runs of the same outcome in real world scenarios (i.e., a
pinball repeatedly branching in one direction, or similar sized planets being
inhabitable) may not appear as unnatural.

FIGURE 2. Differences between contexts in students’ ranking of most likely outcomes in the Equiprobable Ordered Sequences problem.
Compared to previous studies a surprisingly high
number (57%) of students in the COIN/DIE context ranked Outcome #4 ("HTHTH")
among the two most likely outcomes. Thus, in contrast to responses in the
alternative contexts, COIN/DIE students seemed particularly prone to view runs
of a particular outcome (e.g., "HHH") as indications of a nonrandom process and
alternations between heads and tails as indications of a random process. In
sum, differences between contexts indicate that strings of letters representing
sequences of coin flips appear to have symbolic characteristics which evoke specific
situated intuitions—different from intuitions about the most likely paths of a
pinball or the most likely patterns of inhabitable planets.
Reasoning about symbolic representations of random
events in the COIN/DIE version of Equiprobable Ordered Sequence problem may
have been associated with secondary
school-based intuitions. Other evidence for secondary
intuitions was indicated in students' performance on the COIN/DIE version of
the Skewed Probability Samples
problem: Students with stronger science/math backgrounds scored lower than
students with weaker backgrounds (42% vs. 45% correct respectively). Recall
that students were asked to judge the most likely outcome in six rolls of a die
that was painted black on five sides and white on one side. COIN/DIE students'
justifications for judging six blacks most likely often referred to the high
odds for black on each independent
roll. This erroneous focus on independence appeared to be associated with secondary intuitions acquired in
(school-based) contexts.
In summary, comparison of reasoning across contexts
raised some questions about traditional research that suggests specific
universal/innate heuristics and biases in probabilistic reasoning. Some of
these cognitive processes may in fact be tied to intuitions acquired in formal
contexts. Identifying secondary as
well as primary intuitions would help
educators design early learning environments that instill a more rigorous set
of normative conceptual schemata.
1 Normative, in this context, means
"in accordance with mathematical principles of probability."
2 Webster's Collegiate Dictionary
(10th Edition)
REFERENCES
Ahlgren, A., & Garfield, J. (1991).
Analysis of the probability curriculum. In R. Kapadia & M. Borovcnik
(Eds.), Chance encounters: Probability in education, (pp. 107-134).
Norwell, MA.: Kluwer.
Fischbein, E. (1975). The intuitive
sources of probabilistic thinking in children. Dordrect-Holland: Boston.
Kahneman, D., Slovic, P., & Tversky,
A. (Eds.), (1982). Judgment under uncertainty: Heuristics and biases.
Cambridge: Cambridge University Press.
Jones, G.A., Langrall, C. Thornton, C.,
& Mogill, T. (1999). Students’ probabilistic thinking in instruction. Journal
for Research in Mathematics Education, 30(5), 487-519.
NCEE. (1995). Performance Standards:
Vol. 2, Middle School : National Center on Education and the Economy,
Pittsburgh, PA: New Standards.
NCTM. (2000). National Council of
Teachers of Mathematics: Principles and standards for school mathematics. Reston, VA: NCTM.
Phillips, L. D. (1983). A theoretical
perspective on heuristics and biases in probabilistic thinking. In P. Humphreys
(Ed.), Analysing and aiding decisions processes, (pp. 525-543).
Amsterdam: Elsevier (North Holland).
Shaughnessy, J. M. (1992). Research in
probability and statistics: reflections and directions. In D. A. Grouws (Ed.), Handbook
of research on mathematics teaching and learning, (pp. 455-494). New York:
Macmillan.