•

Meatal ComDutatioD and Number Comparison:

• Theil' Roles in the Development of
Sense and Computational Estimation
Judith ThreaJ~1I Sowder
San Diego SWle Uniw:nity
MeOla! computation and computational eSlimation, IWO wpics receiving
increased aucntion (~tionaJ (A>uncil of Teachers of Mathematics, 191W;
Nation·' Science Board Commission, 1982; Pollak, 1986), are presenlly
being adWlCated for inclusion in any newly revamped curriculum. A topic
not receiving such explicit promotion but related to menial computalion
and romputational estimation is tbe ability to compare numbers according
10 their size. Of panicular interest in this chapter is the possibility that more
basi, on t.hcic three topics at appropriate stages in children'S mathe·
matical development will increue what Carpenter and his colleagues
(C"~nter, Coburn, R.cyl, ~nd Wilson, 1976) call "quantitative intuition,"
cw wbat I will call "number IiCnliC,"
'Jlwc Me four constructs thWl far mentioned: mental computation, com·
putational estimation, number comparison, and number sense, all obviously
rclated. What i, Ie" obvious is tbe nature of their relationships and inter·
dependencies. Thii chapter advances a theoretical model of computational
ca'ima1joo ~ attempll 10 explain IOmC of these relationships.
It might be helpful tint 10 diso'u each of the constructs to disllnguish
a",008 them. MenIIJJ cOmpu.llUion is tbe process of carrying out arithmetic
calcul'tions withe",t the aid of external devices. Efficient melllal computing
utilizes different algorithms from those usually associated with
paper-And-pencil caJculations. 1bc objectivc is 10 obtain an exact answer to
lbe c:ompul4llion al problem at hand. Compulaljo!UJl eSlimlJliun, 011 Ihe other
band, doc. not have III ellCi lUlIwcr u ill goal. It it the procc.¥ of con·
¥clliD, from ew~=: LU approximate numben and mentally computing with
IbMe nUlllOt;:lli to obtain an ans'JJocr rcasonably close to the result of an exact
I. 'fhu5, aJthough estimation includes an elemenl of mental com·
il alSl> =;.:~ for appro:cimation, uaing number compamon skills,
182
 I>kn uu Cumpu /"/ lUn "rid Numbu CUmp{UIJUII 1113

,, which I~ our th ird co n~trucl. Number cornparLJon can be defi ned opera lion-
~ lilly. It IS mea~urable by the ability 10 order real num be rs on the ba~is of
,I (ae, such U~ in ~c:IC:ClllIg the hHger o( two or more num ber ~, or by the abili ty
to compare different mllgnitudes, such aM ~electlnl! which o( 1'1010 nu mbe n
i, do. cr 1\1 u limo number.
Nl4l11ba H'nH~, the: hUi! con"ruct, I~ lc:,w c:a~lly 1.lc:lillcl.l bUI relerb to a well
organized conceptuill network thai ellables a per~on to relate number and
operation properties . It can be recognized in the ability to use number
magnitude, relative and absolute, to make qualitative and quantitative judg-
ments necessary (or, but not restricted to , number comparisons, to recognize
unreaM>nable results for calculations, and to use nonalgorithmic forms for
memal computation . A person who has number sense uses flex ible and
creative ways to solve problems involving numbers . The intuitive nature of
numbc:r sc;nse prevents it from being directly taught or easily measured .
The remainder of this chapter first elaborates on mental computation and
numbc:r comparison separately, and argues that devc:loping each skill will
increase numbc!f sc:nse. A diSCUSSIOn of computational c:.timation and its
developmc:nt then (0110'101'. Computational estimation i. ~hown to be a com-
plex of skills and concepts, with mental computation and number
comparison playing major roles. The conclusion of the paper includes impli-
cations for teaching and a list of topics needing further research .

MENTAL COMPUTATION
The mental computation skills advocated by some (Ralston, 1(86) as
ncecisary for numerical competency include the ability to perform whole
number operations on numbers of two or three significant digits, with mul-
tipliers ilnd divisors limited to one signtficant digit. Beyond this, estimation
is ilppropriatc . If an exact answer is necessary, use of a calculator is rec-
ommended. The discussion in this section is therefore restricted to work
with such whole numbers.
Re)e:archers who have worked in the urea of mental computation (Dick-
son, Brown, & Gibson, 1984; Hope, 19H6; Reys, 1(114) claim that the
acquisition of mental computation skills promotes the development of num-
ber understanding by encouraging the search for computational shortcuts
based on number propc!rties. Plunkett (1979) clarified this process through
his IInalysis of mental computation. He: compared methods used for wrillen
(or traditional) algorithms with those created for mental calculations. Writ-
tell algorithms are standardized, contracted (liummuri~ing several SIe:pS
involvinl! Jimibutivit)' lind auociutivity), efticien t, lIutomatic, symbolic,
ge:neral, and analytic (requiring numbers 10 be broken up and digils dealt
with separulely): they arc not easily internaliled (because th!!y do not cor-
respond (0 ways people: nalUrally thillk about numbers), and they encourage
suspende:d under~(anding.
 1&4 N~, COACtpU ,ud OfH1IUJonJ til Ihe MIJJJ~ CruJu

• - •
Mcnual aJgorilhms ha~ an equa.lly interC$ling list of chilrach:risllcs, ~e:v­
eral of whic h arc wo rt h men tioning here with some: accompilnying
. cxplana' ion. (1) 1bcy arc variable. Plunk..ell told of H.O childre:n using 16
difierent me thods to find 83 - 26. (2) They are fle.ubl~ and can be: adaptcd
to suit the numbers mentioned; for example, each probkm, lil - 79,
8J - 51, .nd 83 - 7 wou ld ptobably be mentally c.ilculatcd in a diffcrc:nt
way. (3) They arc act i~, allowi ng the user to choo!Oe a method, whcther
or 00& . (Note the metacogni live aspect of this ~oinl.) (4) They
are hOIisljc in that I hey dul wilh complete numberi ralher Ihan individual
di&i15. (5) Frequently they arc conilrUCtive in the !Oen!Oe Ihal they begin with
the tint number: for example . 37 + 28 is 37. 47.57.67,65. (6) They require
nding throughout. and their lise develops understanding, but they
U('nno( be used to achieve: perfoililance in advance of understanding" (p.
3). (7) 'They otten &i~ an early apptoxil'lUltion of the answer because: the
digits arc calculated first.
Some of Plunkett's examplC$ desenoc a closer analysis. How would some-
ooc nt at mental e&lcu1ation find the differences nOled in (I) and
(2)1 Ptohably. 83 - 79 tW,,1d be solved by aslcing what number added 10
79 &ilia 83. or what number a1ded to 9 gj ve5 13. This reduces to lhe basic
fKl 9 + ? - 13. or 13 - 9. Perhaps unconsciously, In - 7'J is here n!cog-
nized AI the wnc as (70 + 13) - (70 + 9). or (70 - 70) + (13 - 'J).
')bi5. might be ilOf"cd as the rule that if two numben; are less than 10 apan,
their difference Ls a basic facl .
1bc probkm 83 - Sl is lilccly to be thoughl of as (SO + 3) - (50 + 1)
or (SO - SO) + (3 - 1). where 80 - SO is an extension of the basic fact
8 - S. Finally. 83 - 7 'NOuld probably be found by counting down 7 from
83, petillpi in step' of three or four. or by thinking "13 - 7 is 6, so 76."
The lauer method depends on the knowledge thaI 83 is the same as 70 + 13.
Although these solutions outwardly appear quite differen t, they have
some rommonalities. Al least two questions are asked and answered in
obfaining e..ctI solution. C\'Cn though this questioning process might not be
•
a one:
1. How can I upre" the numberi to obtain basic fact ques tions?
2. How will the operational sequence proceed as Ii rt suh of Ihe way I
the numberi1
1b iCC bow these qucstw>ns influence !Oelectw>n of solution processes, con-
Uder lOme wayt of solving 83 - 26. from Plun kett', first poi nt. Some:
plMibilitici include:
(I) 83 + 3 - 26 - 3. or 86 - 26 ji 60 and 60 - 3 ji 57,
(2) 26 + • it 30, 30 + SO it /W, Ilnd IW + 3 i. lB . 4 + 50 + 3 I ~ 50 +
(. + 3) li 57.
(3) (70 + 13) - (20 + 6) is (70 - 20) + (13 - 6). or 57.
 •

UcIlI.rU CUmpu(ullUfl (Mil Numb~f CUmpurlJ()fl 1115

In (I). Ii) 1- ) u~e, Ihe ba,ic litct) r 3. Then, H6 - 26 can be reduced 10

IsO - 20 anJ 6 - b, ~o thc b~lc lact~ II (Icm) - 2 (Ie/l) and 6 - 6 are
u~ed . As a re~ull, Ihe operational ~equence is IWO sublractions:
86 - 26 - 3. In (2), (he problem depends upon the ability 10 complete the
loll<>win~ : tI + ., b 10, 3 (ten ~ ) + '1 (ten) IlM (tl:n ~), iinJ II + '! i) 3. Fuuslly.
Ihe re)ulllng three: numberl IHe IHldell. In (3) , the lacts needed IIrt: 7 - 2
and 13 - 6, wilh two subtractions (ollowed by an ilddltion.
The two tundilmentaJ questions above also influence strategy chOIces for
other optflllions . For example,S x 411 is mentiilly calculaled as 5 x
(40 + II), rc)uliing in 200 + 40, or as S x (50 - 2), re)ulting in 250 - 10.
Here again, numbers are re -e~ pressed in ways that lead to bilsic (act ques-
tions, and the computations carried out as a result include new addition or
subtraction and present new mental calculations calling for a new cycle of
•
questions.
Figure 1 li)b the knowledge and )lcills required to ilnswer each question
successfully. Notice thitt this list contains only topics that are already in the
school curriculum but are often not understood, partly becau)e they have
not been laught within a context in which they becomc personllily useful.
The distributive property, for example, can be used to justify certain SII:pS
in the paper-and-pencil multiplication algorithm, but once the iilKorithm is
leilrnc:d the role of the distributive property is usuitlly forgollen, if it was
evcr undc:rslood .
As the individual r eco gnjze~ mure oplions aVililiible tu answer Question
I, flexibility increases and choices can be made on the basis of speed and
ease . Because there is never just one way to answer Question I, thilt is, to

Quulion I: How cantlie nwntlCtl be: uprcllcd 10 produce

buic {acl

RequiJu 1.1 KJ\()wlc:djt of buic; {tt;1J

1.2 Undi:raWldlnl of pllce value (83 i.IIO ... 3)
1.3 Abilily 10 expreu numbers in di({uenl wIY'
(38 il 40 . 2, or 3S ... 3, or 30 ... 8•... )
1.4 AUIlily 10 ope~1C wilh muluplel and powen of Icn
(UI( 8 - , 10 IOlvc ~O · SO, or •
40 dOO I.... 10 A ) • 100 or h 3a 1000 ... )
•

QIIUBOU 2: What procedural challilea foUow from rc:-<.Prc:uinlllhc numbell7

Requiru: 2.1 Abiliry 10 rc:,:x !tonI Ullin" IJlocialive alld commulllivt
propc:rtic:a 0 ilion ( ISO l' 61· (20 ... 31-150·20) ... (6 - 3) )
2.2 Allilit)' 10 Ule the dlltnbuuvc propen)'
(6. (50 +31-(6. ~O( ... (6 • 31 )
2.3 AbllilY 10 muhlpl)' by po\ItCra of Itn (u above in 1.4)

tilturt I QllcIUllM Guldin. MenIal Compuullion

 19{) N~, COltC~p U and 0 P't'"IIIHU III 11t~ M,J.1I .. (j,u.1CJ

cboosc a form..t for the nllmbers l£l be calculated, mental calculation i!> a
¥tty QntiYe, In\lenti-'C act . 1be child is forced to relale concepts and 5"1115,
thuctw dew'ooinlP • cognitiYe s,"'Clure of how numbers and their opera-
tioo.5 ' . HcDC~ , t..cility with mental computation is a lrutnifeslation
01 number 5eRSe .
'There is oenainly cvKk nce (although admiuedly s~~ ) that menIal com-
putAtion and number underst anding arc c105ely related. In a c ros~-cullural
Iludy of schooled and unschooled O ioula (Ivory Coast) and Amc:ric .. n chil-
dren and adults. Ginsburg, Posocr, and Russell (1981) foun d lilat 70 percent
ollhe ll- and 14-)'Cer-old unschooled Dioula children (contra:.t<:d wi th 34
pel'cent of American children of the same age) used spontaneous regrlluping
to add mentally. For example, the solution proc~ss for 27 + 38
wu characterized as (20 + 7) + (30 + 8). then (20 + 30) + 7 T H, 50 T
1 + (3 + 5), 50 + (7 + 3) + 5, 50 + 10 + S, and finally 65 . (II should
be pointed out ~t ahbotlgh these cbildren had no fmlllal school ing. they
were from a mercantile society, and many had CIperiences in the mark.et-
piece.) In a 5econd litudy (PetillO & G inliburg. 1981), the ability of
Oioula to perforlil mental computation by taking iU.lvanlage of
lbc Itn-, hue and propc:nic:s of the whole number syilem and relationsh ips
betVlleCO operattotu wu Ihown 10 extend to all four o·perations . In contrast,
both lbc Oioula and the American subjects from the first study
reliled heavily 00 school-learned paper-and-pencil algorithms . In fact, the
O»lla, taught in French, were more 5"('«ss(ul calculating in
__ . Iben in lbcir natiYe language, indicating that school-learned tech-
niqucs are 10 IOmC degree contCIt bound.
Hope and Sherrill (1987) studied process'cs used by skillc:d and unskilled
mentAl calculalOrs in grade$ 11 and 12. 1bey fou nd that unskilled students
attempted to callY out roenULI analogues of conventional papcr-and-pencil
algorithms but that ~iUcd students "sed strategies based on number prop-
: ertica IUggClited by the numbers in the taliks. What is not clear from the
Hope and Sherrill Itudy iii how the skilled calculators developed their strat-
egica. Had these ltudenU firit developed sufficien t number sense to lead
the... to bcrome proficient roental calculators? Or did some earlier incentive
to cak'llete roentAlly Lead them to explore numbers and their properties and
tbeD to UiC lbc5e propcnies 10 bccoroe more proficient?

NUMBER
h iI .uronsini to ' " how little attention has been paid to concepts of
Dumber .ize in clemenury and middle schoollcvcl malhema l ic~ tex ts. Erron
made on talk- .~h u ~ntifyini which of two number. i~ clu~cr lu it Ihird
can be calrcmely informative to the telcher by indicating children's mis-
about the Il.Ilture of the numbers they are usi ng procedurally
ill otbu 'kill areu. Understanding a Dumber as i quantity of specific mag-
 Mtnl"/ ('UIII/'Ulw/lUII und Numbtr CumpufLlUfI un

nilUde, more: than, le~s Ihun, or the same as the magnitude of another
number, tS basic to number sense . Of cour!>e, when number comparisons
Ire taught in Ii highly proceduralized manner divorced from number mean-
ings, they lo~ their potemial to develop number meaning in the child and
to provide information on conceptua l development.
The de~clopmc:nt of number magnitude: concepts has nol been widely
Iludtcd . RC::H:ilrch deilting with number magntlude is usually limited in
scope, concentralttlg on one kind of number and/or one age levd, ~uch as
the Behr, Wachsmuch, Post and Lesh (19ti4) teaching eKperiment of order
lind equivillence of fractions to fou rth graders. In un effort to gain il beller
understanding of number comparison abilities of children at diffl!rent ilge
levels and with different kinds of numbers, Sowder and Wheeler (19M6)
asked studl!nt~ ranging from grade 2 through grade 10 to complele twO types
of number comparison tasks . In the first task student~ wen: aslc.ed to choose
the larger or smilller uf two numbers, while in the second they were Ilskl!d
to identify which of two numbers WilS closer to II third number. The results
of this study and of other research studies dl!aling with number magnitude
are most efficiently addressed by considering them in terms of number Iype
(whole, decimal, fractional) rather than suiclly by age h:vel. Within each
domain, modds hu~e been advancl!d or at least suggl!sted for the processes
by which children mak.e number comparisons.
Whole Ntmlbers
Pollrock and SChwanz (19&4) have shown that for adults the ·most viable
model for comparing mullidigit whole numbl!rs with the same number of
digits is the sequential place-value model, in which corrl!sponding digits are
sequentially compared. Children also use this process (Sowder & Wheeler,
19H6). By fourth gradl!, students generally did not have difficulty comparing
number size of whole numbers of four digits and in some instances more
than iour digits . Thl!Y applied the two-step algorithm: If one number has
more digils, it is larger; if Ihey both have the same number of digits, start
at the left and work toward the right onl! digil at a time until a larger digit
is found . This seemed 10 be a self-developed process ralher than one that
had been formally laught to the sludents, and Ihey seemed to recognize that
it could be applied generally.
However, in cases where iu application was less than straightforward,
because ot the large number of digits or because more than IwO numbers
were involved, childrl!n were less successful. For example, only 64 percent
ot the fourth graders tested could tell whelher 43,724,li!9 was closer to
43,624,ld,) or 43,724 ,3H9, whcrC:lI~ percellillge~ corrc:1.:1 were much highl!r
tor 1I11~"H with ftwcr lIigili or only two nLlmbcrs . Thai sel.:ond graders had
not developed ~uch proctdurt~ could be seen by their low iuccess rute
(chance level) on comparing 400 with 368, or choosing belween 24 and 112
as closer to 87 (32 percent correct.) Russell and Ginsburg (1984) obtained
UI8 N~, COIttCIfpU AII4 O~rQ"OIU In the M"Jdl~ G,,,JC.l

&.imiluruulu- uling similar tu h . The difficultie s encountered ~cmed to

be due 00( to alimil.llioo in Lbc fIl(\(jeJ btJt 10 lbe increased memory rc:qulre-
"Kfi" ot woBing with Ih()fe or wgcr nllmbers.

. ticcimal oumb.:rs requires CJ(tending ilnd gc:nc:rillizHlK the

110('00 ot Vtho&e number phce value. Hiebert and Weamc: (19/St)) conjec-
tured that this CJ(tension i.5 difficult because it i, the conceptual fc:a lures of
wbok numbers thAt generalize to decimals and the synlactic feature s that
do not, ,nd children do ON easily make the necessary distinctions. A "more
d}jits ma" bigger" ntLe prompte.c1 almost half of the sixth and seventh '
paden in their 5tudy to select .1814 as the largest of .09, .385, .3, and .1B14.
There are other instances in the literature of children overgeneralizing
blurts of the whole Dllmber sYilem 10 decimal numbers and in turn making
alon when judging Lbc magniwde o( decimal numbers. Sackur-Grisvard
end (.conei'd (1985) found that founh graders compared decimal numbers
(wben the wboLe number pans were equal) by Klecting as smaller the
number wbo6c decimaJ digits CllprCSI the smalLer whole number (or exam-
ple, 3.7 was sek:cted III Le" than 3.53. Although older children (SC:llenth
Vade) were ","ally f'H"ocssful on 'uks such as this, they reverted 10 this
rWc (and other incocre.. t rules) when comparing 5CVCraJ decimal numbers .
1bcsc naLrs often gel corrected as children dc'~elop a better undemanding
01 diCcimel Dumben throogh further work: with them. (n my own sludy
,
bed abow:, cbiJdren "'Cre uked to choose the larger of .00043 lind
_0011. Surocss rates at gradell 6,8, and 10 increased (rom 13 percent 10 53
__ . . to 9S DCrtlGot. When ISke.c1 wtuch was clOKr to 39.42, 394.1 or
58.70, f'IQOCU al the ume three grade klt'ds increased from 55 pc:rcem to
78 percent to 95 percent.
There is no need, however, for Ihis lengthy self-correclion process.
Wc.arDe and Hiebert (thi' W>lumc) dcaaibe a leaching IICquenec: (or instruc-
tioo 00 decime! Dumbers. Fourth graders comple~ing this sequence were
abk to choose the Larger of two decimal numbers by a semantic rather than
a syntactic process, eYen though such task.s ha.c1·OOI been included in Ihe
•
•

F,aaions
la the Sowder and WheeLer study (1986), s"cee" al idenlifying Ihe larger
ot two fracfWlru aJ.so increased ~r grade levels: from LO percent to 19
to 37 percent to 95 percent over grades 4, 6, 8, and LO when com-
paring 516 and 5/9, for example. 1lw: extremely low 6c~res before grade 10,
area lhoulh .COCCI incrc:uc wilh Ilge, ill'licllk thltt children lind (ractional
CIODCCptI even more difficult than declrrutJ COncepti. Children do nol litem
10 attach meaning to tractio/U beyond that of a fraction III "pan o( a whole,"
&ad their oroocdures for working with fractiolU indicate thaI they prefer 10
 ),kllluJ CvlllPU/U/IUfI ufII,l Nurn btf c;o"'P""o un IIS~

creat the lIumera tor and the dcMrninilt(j r iI~ two ~r<l r iltc cntlflt:S rilther
Ihal1 to accept a (rac fl on all a number in It~ own ri ght. FraclJun compaflwn
i.!. rruide more difficult by the posblbility that ne ither of two numbers I~
t;,rger, even though the numbers ilppeilr to be different.
8ecau~ eq uivalent (or equal) trac tion~ are so imponilnt in computational
work, cotnpanllg fractions has received )omewhilt morl! rc)carch attentiun
than comparing dc:clmal numbc:n (Behe et aI., lYt:!4 ; Kchlakl:, IY!sO ; Peck
& Jencks, l~IH) . Foe example, Peck and Jencks (19t:!1), working with sixth
graders, found several who claimed that 2/3 alld 3/4 were I:qual becau~
"there were the same number of pie ce ~ left over" (p . 344) . Referring back
to models 10 malc.e comparison decisions can cause additional problems
bcCilU~, as they noted, children' are "mode l poor," with many having a
circular regIon as their only model of a fraction. Funhermore, understand-
ing of order and equivalence of (ractions is dominated by knowledge of
ordering whole numbers (8eht et aI., 1984) .
The comments made earlier regarding the slow devdopment of decimal
number ,ize concepts are equally true for fractional number size concepts.
1ne usual curriculum is at fault in allowing understanding of fraction num-
ber Kile to develop so slowly. The lack of such understanding is particularly
detrimental because it hinden other work with decimal and fractional num-
bers, in some ca~s leading students 10 memorize algorithmic rules as their
only recourse. i(erslak.e (1986) concluded, after teaching experiments with
13- and 14- year-aids, first teaching a mOdule 011 fraction concepts herself
and then with six volunteer teachers, that the use of a variety of instances
of equivalence incrt:ased children's ability to deal with tasks involving equiv-
-
alcncc. Instruction with children as young as those in fourth grade can
diminish whole number dominance and lead 10 improvement on ordering
fractions (Behr et aI., 1984). The important implication from both studies
is that more instructional time on establishing meanings for fractions can
be wurth while. The next step is to link these meunings to computational
procedures and to accomplish Ihis within regular classroom settings. This
ill a fruitful area {or future research .
The connection between the comparison of fractions and development of
number sense is dear. Comparing fractions is necessary for obtaining an
intuitive feel of the size of fcactiom . If II fractional number is recognized to
be! close to 1/3 or 112, for example, one has a beller feel for its magnitude.
111i& fractional number sense ial particularly important when estimating with
fractions.

COMPUTATrONAL ESTIMATION
COntiider the following problem: Jllek'~ car gCb 2~ miles to the gUllun.
The gas tank holds 14 gallons. It it is full, about how far can Jack travel
befote he has to SlOp for more glls"
 190 • N~, COftCtp U and OptrlUIOJU in Ihe MIddle Grudes
•

-
One way to 50Ive thii problem ~lIOu1d be to round 29 to 30 14 10 15 and
the ' ,
.. mcn~y compute 30 X 15. CompuLaI~nal estimation, then, involves
~ IUblkiUs: apprOlimatk>n and mental computation. Skill at approximat-
ana de.pcn~s upon the ability to compa re n u mbers. Allhough the
approxlmAt1na done in the Cllampll! problem ~ems 10 be a case of simple
roundina, it is not 50 "mpk: as it hrst appears. Rounding, liS it is frequently
UU&N in Ichools, is a purely procedural skill in which "you round down if
the number ends in 1 through. and up if it ends in 5 through 9." Many
deUlentary school $ludenu would round 14 to 10 rather than to IS in this
probkm.
Inability to make judgments of relative size when approximating numbers
is detrimental to woccss at computational estimation. In a study (Thread-
•
gill-Sowder, 1984) in which middle school age children were asked to do
Of' NAEP-type computational estimation t as~s and tell how they
L __ ' _ their answers, a large number of errors could be attributed to lack
- -'

of understanding of Dumber size,leading to poor approximations. For exam-

p&e, wbeo $ludents were asked to estimate the sum of 148.72 and 51.351, 35
pc:n:(Ul of the 29 $ludents intenoiewed rounded the 148 and 51 10 150 and
SO and added; they lhe.n returned to tbe decimal portions and rounded and
added 'nin. ~og to a final estimate containing a decimal portion . These
$ludenu were unab&e to recognize that tbe decimal portions of the numbers
v.cre 'melier than the errors introduced in the initial rounding and therefore
inc.illoi6caol. 10 anotber problem, students 'h'Cre asked to estimate 7M\) x
0.52. Only 19 percent IIsed 0.52 as 0.5 or 112 or 50%. Several students
.52 to either 1 (llsing scboo!-uught rounding rules) or 0 (because
"if' very .mell"), leading to IUliwcn of 789, SOO, or O.
menl.a..l computation and approKimating using number comparison
are Mceuery componenu of compuutional estiflUllion, the concepts and
.kills discus$Cd eart~r as prerequisites of these two components are also
prereqllicilCS of compuutionaJ estirnatioo : knowledge of basic facts , under-
uanding of number operations, undentanding of place value, ability 10 work
with multiples and ptYol'Cn of 10, and understanding of the symbol systems
used fOf' decimal and fractional numben. Other resea rchers have also found
that c:ompullukNW estimlltion ability is related to theiC number concepts
and .kill •. Rubenstein (1985) examined the performance of eight h graders
00 four computational estimatioo scales and on eight related sca les. She
found that operating with powers of 10, judging the relative size of numbers,
and comparing itenu filch as 39 x 29 with 29 x 41 in terms of si2e all
cnp1ribuled to eujmation performance. Likewise, R.eys, Rybolt , Bestgen
and Wyau (1982) '~riCleri7c:d good clliJrull.On (~enlh gradcrb 10 a d ull~ )
II havina • lOIid IeRIC o( plMce vaJuc, tbe ability to reformulate numerical
der', s.lr.iU at menl.a..l computation, a command of arithmetic properties, and
to a variety of wateg~' that changed depending on the nu mbers and
opcratiooa inwl\'ed.
 .,
Mtllta / CompUtutlO1l unJ NUIIlbtr C()mpurl.!()n IY !
•

R.eY' Cl al. (11)8 2) also found Ihal good estimator) conllis tently used three
r basic proces~ s . The first. reformulation, calls for replacing Ihe original
numbers with new Ones before computing . Examples of reformulat io n
include lruncalion ("chopping off' the right -most digits). rounding. aver-
aging (noticing thai several addends are all close to one number and usi ng
Ihut number 10 compute) . lind chllnging the form for a number (for example .
(rom 112 to 0.5) . Notice that this process incorporales the approx imat ion
skill described earlier and is heavily dcpcndent on the ability to compare
numberi . The~e changes, particularly trunceation, often ceal! for an adJu st-
ment in the eSlinlale through the second process, called compensation . The
third process, translation, takes place when the estimation process brings
about a change in the structure of the problem. For example. rather than a
multiplication followed by a division. the mental computation might be
eallier if division is followed by a multiplication. This process is used 10
make the resulting mental computation easier.
Consideration of the work of children JUSt beginning to estimate also led
to a bette!r undemanding of the (actors involved in the estimation process .
Interviews (Sowder & Wheeler, 1986) with children in grades 2. 4. 6, ti, and
. 10 indicated that children in grades 2 and 4 were generally unsuccessful on
estimation tasks. By sixth grade, students exhibited a bella understanding
o( the goal of a computational estimation lask but continued to be lIery
hesitant hnd uncerlain as to how to go about reaching that goal. Not until
grade 10 did students seem to comprehend immediately the nature of a
computational task. and exhibit confidence during the solution process.
If became clear during these interviews thai lack of success on the esti-
mation task.s is not simply an inability to carry out the component processes.
The students who could not estimule did not understand the naturc of the
computational task . They did not recognize that it was appropriate to
approximLite before computing ralher than after, that the final outcome was
to be an approximation of the answer that would be obtained by straight
computation. that the needed accuracy of the approximation depended on
the problem context. or that it WIU. possible to obtain different but equally
good estimates by different but equally good processes.
In Figure 2 r have allempted to delineate and rdate the vurious compo-
nents of computational estimation .
The dependence of computational estimation on mental computation and
number comparison skills was explored in a study of the development of
computational eSlimation ability (Case & Sowder, 19H7). This study was
based 011 Case's (1985) nco-Piagetian theory of intellectual development.
The theory propo'cs thaI children's co~njtjve ~rowth proceeds throu~h two
MtUgcli durinllihe IIchool yeatli. The dimtllsiofluJ sla~e (approximulely uges
5-10) is characterized by the number of unitli (or dimensions) children locus
on at any Olle time: one, two, or multiple ullits during the three substages.
When children can focus 011 and inte~rate two or more units they puss into
 Primary Skills Required SpecifIC
- Durin, Cumpul.ltiaNI FIOO
(Reya et Il.. 1982)

'ft),. AppoWMlt
wtaak 8I'C u.JCd lID en ,I
(Chan,e nwnben 10 •
~m ..

•, wbol

Callpensa t ion

Transl8 !ioo
(Olanae snucture 0[
plOblun for ease of
menial OOllputation)

Mcwethanc.Lr

F,gur<] Component> 01 Computational Estimation

 •
•
Mcn/gJ COlf'PUIUIIO" urid Number Comparl.lUfI
193

the IItClorU/J ~taue (. ' . . .

• .' r> ilpprOXtmately IIges 11 -18). dUring whtch they become
II\CrC4lslOgly capable of coordinating qualitatively different dimensional sys-
tem, . The two componenu of computa tional estimation mental
compul~lion and converting from exaCI to approximate units usin~ numbe:r
complUlson I>kills, satisfy the criteria of two qualitativety dif(c:rent systems .
It therefore follow£ thaI children should nOI be able 10 coordinate the~e
components within a computalional tstimation task until thc:y are in the
ycctoriltl stage .
For each of the three: substages within each stagt, tasks were designed
lhal were predicted to bt stage appropriate . Within the dimensional stage.
Ihe tasks demanded succc:ssively more difficult mental computations and
number comparisons, with some very easy computational estimation tasks
al the last subslage . Wilhin the vectorial stage, the computational estima-
tion ta~k~ al~o became more difficult at each successive subslage. Students
at grade levels K, 2, 4, 7, 9, lind 11112, considered to be representative of
the six substages. were given tasks deemed appropriale for their level and
tasks considered appropriate al the nexl level.
Generally speaking, task placement as predicted by Case's model was
remarkably accurate . This developmental model of menIal computation,
number nearness, and computational estimation predicted performance
ba.ed on conceptual development rllther than on school-based procedural
learning. This was possible because tasks such as the ones used in the study
were not included in the curriculum at the time, so the children had not
been exposed to explicitly taught algorithms for estimation . This study has
implications for curriculum pllsnning, particularly in light of the current
movement to include computational estimation in the elementary school
curriculum.
The: development of the concepts and skills specific to estimation as spec-
ified in Figure 2 was the focus of yet anothe:r study (Sowder & Wheeler. in
pre"). Besides the: usual direct-answer computlltional estimation task, stu-
dents were preserlled with estimation problems together with solutions of
hypothetical peers. One of the given solutions exhibited an understanding
of the concept or use of the process in question. while one or more other
solutions did nol. Students were aslu:d questions that requin:d them to
contraSI and compare the different S9lution strategies and answers. Problem
tasks designed for grades 3, 5, 7, Ilnd 9, became increasingly difficult at
successive grade levels.
Except for a few third graders, in general students at allugc levels tested
jII'upted multiple: processes for obtaining an estimilte . However, only about
hili( ot lhe: ,Iudelll. lit grlldell 7 lind 9 t;nd Cltr fewer Itl ~radc~ :\ lind 5 wen:
willing 10 con~ider different hnlll e~timlltc:& el.jultlly ucceptllblc. This probu-
bly reflect. their familiarity with It iChool curriculum where only one Itnswe:r
is correct.
In terms of conceptual understanding of compulUtional estimation, per-
J~ NwrtiH, COItCC'pu aNi OpC'fGUOtU In IhC' MIddle Crudes

~ tbe ~ interesting respon$Cs in this iludy were given during tasks

req uiring compari500 of answcl'$ to problems where one hypotht:llcal stu-
dent aDDro.im.~d then computed, while another student computed then
, boc.h geuing the 5ame answe r_ The expectat ion was tha t
DaCr students would reject the approximate-then-rompute strategy, and
o&der students would rejcct the: compute-then-approximate strategy. In fact,
ju5t the oppo6ite was true _ Altbaugh most students accepted the use of
. - numbers and the resw, of comput ing with them, thl! num ber
ol 5ludi:nts stating that appro.imating then computing was preferable to
computing then approximating decreased with grade lcvc:l .
What is panicl' lerly inlere5ling aboutlhis resull is thai we see an example
ol students who by ~enth grade demon5lra te the ability to carry out a
,eoeral procedure, largely $Clf-taught but apparently found to be efficient
and therefore advantageolls. They leayc us with the impression that because
they baYC disconoc:red and lise this procedure, they have achievl!d a cenain
~ 01 conceptual understanding of the computational estimation process .
Yet in the: final analysis, they are uncomfona ble with this process and
it inferior to a process that yields an exact answer.
Of the estimation proceueli listed in Figure 2, compensation seems to be
ID05t cloiely linkCL! to conceptual undeflitanding of computational estima-
tion ,nd is therefore of paniCular interest. Thili skill is also untaught (at
to the students in the study Oc:scJibed here), and its use may indicate
aD ,warellCu of Ilk: need to consider the degree of error between an est i-
II'UlLed and aD cuct am'wer and to make adjustments depending upon the
probkm $CIting ,nd intended "lie of the estima te. Fifth graders, when com-
paring a probk m in which a gron es'imate is follO\lJ'cd by some son of
(ompenntion to a problem without compen59tion, choose lili: compensated
,ns""," as better, whereas third graders tend to go either way. 11 is, however,
__ unlil ninth grade thai students are likely to compensale in their own
work, sucb as adding an exira dol/ar to $32 when est imAting 4 x $8.27.
_ __ it could cel1einly be argued that $32 is a legitimate estimate, the
~oinl i.5 thaI it did nol appe.a r to occur to the sevent h graders 10 go beyond
$32, ~n after 5CCing earlier examples whe re litis type of compensation had
l,ken place. Ninth graders are more likely (wi thou t prompting) to continue
with the problem through a compensation phase .
Edwards (1984), after analyzing and dism issing the o ft -suggested reason
for teaChing computational estimation skills , tha t is, so tha t errors made in
a calculator could be recognized, reasoned tha t the fu nd~menlal jus-
for teaChing computational estimation is tha t it develops number
sense_ As be defined iI, number sense includes mental ari thmetic and some
10ft 0( cepacity to compare numben . Hii claim introd uces the question of
which COllie' fint, number $Cnse or computational estimation? Rather than
atlemptinj to JUcc • "riel IUerarchicel structure on the devc:lupment of
lbeK two COlUmna, or for that mauer, all four constructs, it seems more
 . -
/rIm ,ul Cu ,IIPu/UI' UII " fill Number ('UtrlpurlSufI IIJ5

rea ~ ona bl c to cons ider each ala feeding on lind strengthening the others in II
~pi ra l fa ~llI o n , with computational estimation entering the pic ture slightly
later than ea rly stages of mental computation and number comparison abil-
ities, accompa/lled by an increasing number sense .
•

CONCLUSION
The: relallonships among mental computation, number comparison, and
computational estimation can be found in their common heritage of depend-
ence on place value concepts, on the ability to work. with powers of 10 and
to use propenies of number operations, and on the understanding of the
symbol systems used to represent numbers. The development of these skills
(orces one to explore the role of the prerequisite concepts and skills and in
so doing connect them to a cognitive structure that is here called number
sense . This structure continually changes and extl!nds to incorporate new
knowledge as the practice of these skills brings about the discovery of new
relationshi ps.
At the Research Agenda Conference on Effective Mathematics Teaching,
Peterson (11J1J7) presented a paper on teaching for higher order thinking in
mathematics. (A revised version of the puper appears in a companion vol-
ume .) According to Peterson, recent research and theory suggest thr~~
processes that facilitate higher order learning in malhc:matics: "(a) II focus
on rru:aniog aDd understanding of the learning task, (b) encouragement of
studem autonomy, independence, self-direction and persistence in learning,
and (c) teaching of higher-level executive processes and strate!gies" (p. 45).
The type of Ic:arning we have been discussing in this paper is certainly higher
order learning, and the processes listed by Peterson offer some guidance in
thinking aboul (he issue of instruction. Both Peterson (1987) and Doyle
(J983) recommend indirect instruction to bring about the changes demanded
by higher order thinking. According to Doyle, i.!,\direct instru~t! .
carc:fully slructure 0 rlunities for (f ' . ..
L _-.. . 'h direct and ample experiences wjlh cootrOl.
----

This brings us to the first set of research queslions, all dealing with
imtruction . Can effective teaching methods be developed for mental com-
putation, number comparison, and computational estimation that will lead
to incn:al)ed number sense'! How should these topics be sequencl!d'/ What
etliclency level:. arc appropriate and reasonable? How can We! best monitor
student performance and growth? What effect does long-term instruction in
these topics have 011 the development of rational number concepts and
skills? And what effect does it have on calculator usage and on the use of
wrillen algorithms?
Another rCl>can:h question cOllcerns the current movcmenltO incorporate
algorilhmic thinking into the curriculum at all Itvels (Steen, 1986). Algo-
rithmic thinking is used here in the "contemporary" sense, referring to Ihe
 J96
•

aUljoo of aJpilbms and I.bc mind iet Ollhin.k.ing in ~rrll' of .. Igonlhms

wbca (Maurer. l~). How do tnCnla.l computation and
. fit in here? For ~ (if 00( tJl) problems of tht~
there as DO iel way of pr0C«4ing. and in most i"'Ia~, there is nOl
• .. ~" way. 1bc question 01 the effea that the learning of mental com.
and . C5tiaatioo can lutvc on the la~r developnu:nt
of thiokin& ill.bcreCore · inlerening.
Lut but noc arc I.bc questions raUed by the model of computational
put forth io ttW paper. 1bc model iJ Itill sutchy, and the cog·
ll"dcrlying each of the COfl$tmcu need tD be more cartfully
10 that pre4irtiom about task perfollnenu can be made and
•
. tw ooly recently been 'tuilied, and there
as )Ie' lO be learned. Rcys (penonal communication) has suggested
do DO( apprccia~ the pocentitJ poY;'cr of eilimluion, par·
as • unifyiO& lbcmc throughout the 51udy oC rational numbers .
Pub.p&. theory oC the coanitivc peoce"CI underlying eSli·
matioo ..... "d not only what is required in learning tD euimale but
a1io ..... ·Id darity iu P'~ in the curricuJum and the degree of 1I11enlion it

RE.f'E1lf NCF S
BdIJ. ... J., Wacblawlb, 1.. lUI. T. R .• It. R. (1984) . Order and equivalence o( rallonlll
-"cIa: A diKaI te"';:~;"11 JowMJ jx R.tuGtcJt ill ~JnlUia Ed.. callon.
lJ,l2l-14I.
C",-I". T.P.. O:lbun" T,G. , Rty'. R.E, ... WHwo. l ,W (1976) , Nocc. (rom NaIiOnaJ
. FA i'a I jon . ArIiJanalIic TMdwT, 2J. 296-302.
C-. I.. (ISIIS). JllJAl«mo' III. 0riaDd0. fL: MI+mic Pr". ,
C-. I.., .. J. T. (1987) . 1M lk..dopMelll of comp.... GorA' WUnelU>n : A n~o-
,........ ~ , . 10 be ,ubcnillcd lot publialioo. .
~. L. Browa. 104 , ... G1b.on. 0 , (1984) , ~ tnllIlvnuwcs: A uucnu'.
f'tlI'IM 10 reu:al1'f:IIM'dt . Fe., Sana. Fnaland: Ho". Rindw1.nd WiMlOO .
Do,tc. W (I~J) . .f:gAcmic WOI'k . RI~ilw of FduCeDNW R.tuGtcJt. 5J. 159-1\19.
~ A . (l984). ea'im"Q b N"'''racy. Fd'ICQliOllDl SludltJ JJI M.iJl/K •
..-ia, lJ, S9-71.
GIMhur,. r.,
H. ~, J. K .•• puncH. P. l. (19111) . Tbc deoclopmcnl of mcnl~1 add'llOf\
_. ~ al.,IKIOIina ud DillllfC . JowJttJJ of Crou·C..ulUili PsycJtoJogy. 12. 163·1711.
tWbcil. J . .t YIlurK. D . 1986). Plocedures OOJcr COOlCCpI$: Tbc acqui~llion of Ikcuru,1
. ID J, (Ed ,). C~ptWJ AItd p'oudlUili ICI'O...kd~,· TIv C4U of
(pp. 203-22)) , Hilltdllc. NJ : LawtCI"X Erlbeum AnoriILCS ,
11npc. J . A. (1986). Mcnlll nkul&lion: AMchroo i'm Of balic ""ill? tn H, l. Schoen &I M. J.
Zooeaa (Eda.), £uiM9liDIt ..d IfVnlp' ' (pp. 4S-S4). R.ca1Oll. VA: ruliooal Coun·
cil 01 Tea' ",i 01 Malheh"'ia.
I . , J. A .• .t ,J , 104 , (1981). Cbaractcri"ics 01 un~kil1cd and ""illed mcRlIII Clllcula·
Ir.I. JowM/ jx RaMleJa ill M.eUtt"lQljc. £4ucP'Wn. 18, 98-111.
• D. , factioN: CJtjJJrtn', 'JralCll1U IINI CmNJ . WiMIOI. BcrlUiirc. Enlll"lId :
NfER·NEl.S()\/.
Wawrr, S. 8 . (1984,. n.o oIllpilhmic m· .bemalics. MiuNmQJia ullCnu, 77. 43U-
ill,
NaUauI eou.aI alb'hm 01 toUlhcmalia. (1~) . An lJ~ndiI for ecUDII: /UcomnunJilllolU
p, IdtnftI . of 1M J980',. DC'lOII. VA: .""hor.
 lY7

HelMe) Sc;cnce &.rJ C(){(ln",I ICHl ; IUpon ,,/ ItlC CoolettllC.< fi.o.1Cl of the M.It.un. IIC. 1
SaotDCa . (1~2J . TM fftlJl~M4J1CIIJ 'OC)I« CWIIC WIUJI X-l1: WNuU U"I/~fl/41 aNI
~ u ,.Of. Wulu I\~lon, DC: N.lt ion, t Sauu Fou~ 1100.
ftck: . D . M . .It knclu,~ . M. (l~tlIJ . U>ncCPhlal iUI>CI In ItiC leechJn& iinel Ic:.r wng of 1r;.'lIoo~ .
J"."IUJ frx Ruulfch In MDIMmOJIC.J F.d'lclUioJl , 11 , HIJ-J.48 .
ft~uon, P. L . (1IJH7.M~rchJ UIlChlll1 frx JlIgM,-wdtllillfl)ung In mwllufrIJwCl The cJa.dkl'lllt
jew 1111 MAl dtZ'ilcJl . Po per pre~nled ,t ~W:ilch Agenda Project Meelina 00 EltcClI~C
M.iltiCmMlica TellChullI, ColumbIa, MO .
Pchl~, A . L., '" Gin.burg, H. Po (19111) . Menllil ifllhmcllc III Afr.c" 000 Amcrx;.a : Slfilc~e"
prIOClpkl, 101\0 upl~n.tiom . Inltrnul/Ofl4/ Jou"I{~IIJ/ PJyclwlugy, 17, 11I -IU2.
fwn"", S. (InlJ) . l)ecomf>V\lIlOn IoU alllhill rot. MDlhmwui("J 11'1 StilOulJ , II , 2-S.
PoIliit . H. (191lob) . n~ 1C1IOu1,fl4llhe11\Q1IU cumcu/um . RilUl/ig nUl/unu/ t.lp~ClU/jUru. Jumnuuy
of COfl~'t"U . W.. ullnillon, DC: MliltICmaliclil Sciel\('~1 EducallOn Ua,ltd .
PoIUlICk, S. E .• & Schw.. m, D. R. (IIJ~) . Complitilive judgA:menil 01 mulildlgll numbefl .
Jry'TIwJ of £..cpmmtntul PJychol0BY : uarning. Memory and CognillOn, 10, 32-'15 .
Ralilon, A . (11J1I6) . DIU/I upu" ullM uuJc foTct on cutrleu/um /flm~wo,Ju for X-12 fftlJllu-
1NU/c1 . WllWIlnlllOll, DC: MIII~malicAI Sc:ieflU' EducllliDn BDII/d.
RtYI. R . E . (11j1j4) . Mental cumpulation and ulilTUllion : Put, prcKnlilJld fulurc . EkrrunuITY
SclutOJ Journul. 1l4, 5<17-557.
RtYI, R . E., Ryboll, J . E. Elt~I~C:Il, 8 . J., & WYIIIl, J. w. (IYII2) . Prtx:CI~~ u..cd hy ~oud
cOlllput.lioruAl c.lim~IU(l . journal frx JUJtjjfrlt in MoiIJufftlJiIC' £duaw"", IJ. 111)·201.
lWbc:nllcin, R. N . (IYHS). CompullillOfUl cllimlulon Imd rclaled tnllltIClTUlliul Ilul1l . Jo..r1WJ
p RlllaTch III Mal/~f"f\IJl/CJ £dW:U/jlHl, 16, 106-1111 .
, R. L.. & GimburK, H. P. (III~) . CoKnihV\: IIMly,il oC .:hildtc:n'~ nlllltiClTUIlLct dilli-
cWlacl. COllflilion u"J IrulruelilHl, J. 117-175. .
SKlt.ur-Gri."ard, C . & u:onard, f. (1985). Inlc:rmc:lIlille OOtlnllivl: organilat lun~ an the process
oC lcarninllil malhcmaltcal cDncc:pl: 1bc: order 0.1 posilive decinUlI numbers . Cognilion and
lfl,S/ructioll. 2. 151- 174 .
Sowder, J . T., & Whc:c:lc:r, M. M. (191Ul) . Thl dtw:Jopmenl of compultlllonul tSlflfI4IliOIl Ilnd
IUUfIMr It'Ut: ,",0 tApioralory m~ditJ . S&n Dicl!o SllIle: Uni'le:rsIlY: CRMSE Repon.
Sowder, J. T., &. Whc,l,r, M. M. (in pre") . The dt;vc:lopmenl DC .:oncc:p" IInli ,wlleglC:1 used
in COIIlp41 III Ii<lnlll c:»1 imilliu/l. JournQI for Restareh ill MulhtmJll;cS £duelllwn .
Steen. L. A . (I~). olCes ful clulngt ill. 'he K-J.2 m'dhtlTlU/f~S eurri~·ululII . Paper prepared lor
liac Conlcrcnc, on School MattICmlillCl CUrriculum: RIIUInI! Nalton,,1 ElpeclllllOm .
TIucadgiU-Sowde:r, J. (11JK4) . Compullilional cllimllli<ln procc:duru ol IiChool children . .Joomul
0/ £4f1CUliOfUJI RlJturch. n. 332-))5.
ACKNOWI.EDGEMENTS
TIac work reponed hue: WiU putililly runded by Ihe: Nlllional .ScIl:ncc Fou~alion under
Oranl OPE 1l320211H. Any oplnioOi, tinJmg_, lind cOl\Clu~iom or. rc:com~cmJllllon~ lirc lhose:
III IbG aUlbcm liod do nol nc:a;cuarily rcReel the vie:wl or the: NJilionil Sclc:m;c Foundlillon,