Models For Predicting Anthropometric Dimensions of Students Needed For Ergonomic School Furniture Design

Models for predicting anthropometric dimensions of students needed for ergonomic
school furniture design
S.O. Oladapoa; O.G. Akanbib.

a
Department of Industrial and Production Engineering,
University of Ibadan, Ibadan, Nigeria.
+23408060530526
dapobola@gmail.com
b
Department of Industrial and Production Engineering,
University of Ibadan, Ibadan, Nigeria.
+23408066266848
engrakanbi@yahoo.com
S.O. Oladapo is the corresponding author and can be contacted at:
dapobola@gmail.com, phone number: (+234)08060530526
Abstract
The study aims at developing some models that will make use of easy-to-measure anthropometric
dimensions to predict difficult-to-measure, time-consuming and energy-sharping dimensions needed for ergonomic
furniture design. A total of 160 students aged 14 to 16 years were randomly selected from eight public secondary
schools in Ogbomoso, South West, Nigeria. Exhaustive search methodology, implemented in ANFIS was employed
for selection of the needed inputs out of the five measurements (stature, waist height, shoulder-grip length, lower-
arm length and shoulder breadth). The selected inputs were analyzed using response surface methodology,
implemented in design expert software. The best models are ones with no/non-significant lack of fit and highest
coefficient of determinations. The study revealed that 11 out of the 12 developed models exhibited non-linear
relationship. This contradicts the long standing assumption that the relationship among length dimensions is linear.
The furniture industry will benefit from these economical, adequate and effective predictive tools.
*Corresponding author. Email: dapobola@gmail.com

Relevance to human factors/ergonomics theory
The paper presents research models on evaluation or prediction of several anthropomorphic features which
will allow the design of school furniture in optimal way. This will increase productivity in term of students’
performance. Also, musculoskeletal disorder and injuries among students will be reduced if not eliminated. It will
also lead to reduction in production cost on the path of government, especially those of developing countries,
saddled with this responsibility. The furniture industries will benefit from the research as their product will meet
ergonomic standard at affordable cost. Furthermore, inventory cost on the part of the manufacturers will be greatly
reduced if not eliminated.
Key words: furniture, students, dimensions, models, school, design.
1. Introduction.
Chairs and desks are designed with the intention that they will provide comfort to the
users (Chung and Wong, 2007; Jeong and Park, 1990; Turnay and Melemez, 2008). Another less
noticed function of school chairs and desks is to ensure children stay in one place in other to
facilitate learning, monitoring of behavior and performance and to minimize distracting
interactions (Knight and Noyes, 1999).
Bad design of furniture may lead to health and learning problems. For example, learning
is affected since uncomfortable and awkward body postures can impair the student learning
interest, even during the most stimulating and interesting lessons (Sarni, et al., 2007). Anatomo-
functional issues are another consequence of the mismatch, where postural changes, general
tiredness, psychological manifestations and musculoskeletal dysfunction can increase during
high growth stages such as puberty, (Savanur, et al., 2007). Lower back pain noticed among
taller adolescent students can be attributed to various factors, being perhaps one of the most
relevant, the mismatch between children anthropometry and furniture’s dimensions (Grimmer
and Williams, 2000). Despite these challenges, studies conducted in several countries

surprisingly revealed that there are high percentage of mismatch between school furniture
dimensions and students’ anthropometry (Agha, 2010; Castellucci, et ., 2010; Dianat, et al.,
2013; Gouvali and Boudolos, 2006; Mokdad and Al-Ansari, 2009; Niekerk, et al., 2013; Sarni, et
al., 2007; Savanur, et al., 2007). This has been attributed to the fact that school furniture design
is typically based on decisions not related to the students’ needs (Barli, et al., 2006; Corlett,2006;
Parcells, et al., 1999)..
For this to be remedied, there is a need for complete anthropometric data for each country
(Garcia-Acosta and Lange-Morales, 2007) and based on anthropometric data obtained from the
intended users, every country can design fitting furniture for school children (Molenbroek and
Ramaekers, 1996; Oyewole, et al., 2010) instead of a one-size-fits-all philosophy that has been
adopted in the industry. This may not be unconnected to the fact that measuring and gathering
anthropometric dimensions needed for ergonomic furniture design demand a lot of resources in
term of time, workforce, equipment and funds. It is unlikely that the furniture industry will be
able to rise to these challenges without better tools that can help it to achieve the desired
furniture at minimum labour and time. Hence, the aim of this study was to develop some models
that will make use of easy-to-measure students’ body dimensions to predict difficult, time-
consuming and energy-sharping anthropometric dimensions needed for ergonomic school
furniture design.
2. Material and method.
2.1 Study design.
A survey was conducted between July and August, 2014 to measure students’
anthropometric characteristics in eight selected high schools in Ogbomoso, Oyo state, Nigeria.
Four schools were selected from each of the two Local Government Areas (North and South)

that formed the nucleous of the city. This was done to have adequate spread and to obtain a
representative sample of secondary school students in the area.
2.2 Participant Sample.
A total of 160 students (80 male, 80 female) of those who volunteered to participate in
the study were randomly selected from junior secondary school three to senior secondary school
one. Their age range is between 14 and 16years. They had not participated in such previous
study and had no physical disabilities. This is done because Jeong and Park (1990) had stated
that Sex difference in anthropometry is significant for school furniture design. Furthermore, the
phenomenon of variations in body proportions among genders, ages and requirement of suitable
furniture was also reported by Chung and Wong (2007).
2.3 Statistical power analysis for the present study
The adequacy of the sample size was verified using GPower 3.1 software version. The
significant level and effect size that were used are 0.05 and 0.1 (small effect) respectively. The
result of the analysis; sample size 160, number of predictors 4; returned a Power = 0.9753871.
Since a power of 0.8 or greater is considered powerful conventionally, then the result of power
analysis of 0.9753871 is adequately sufficient. However, if too many observations are used, even
a trivial effect will be mistakenly detected as a significant one (High, 2000). By contrast, if too
few observations are used, a hypothesis test will be weak and less convincing. Accordingly, there
may be little chance to detect a meaningful effect even when it exists there. This led to a priori
analysis to determine whether the sample size was actually not below or above the needed
number of observation. A power of 95% was employed and the analysis returned a sample size

of 132 as adequate. This justified the sample size of 160 used by the researchers as the resulted
difference meant more power for the research.
2.4 Description of anthropometric measurements
Measurements were carried out on the right-hand side of the participating students, while
the subjects were wearing light clothing and were barefooted.
The students were required to sit in such a way that their thighs were in full contact with the seat,
their lower and upper legs were at right angles, their feet were placed firm on the floor surface or
on the wooden pieces under their feet for the following measurements; shoulder breadth, knee
height, elbow height, popliteal height, shoulder height, buttock popliteal length, shoulder breadth
and hip width. They were instructed to make use of the backrest such that the back and backrest
were in good contact and their trunk was upright.
For measurements that required the subjects to stand such as; stature, waist height,
shoulder-grip length and lower-arm length; they were to do so while standing fully erect with
both feet together and the head orientated in the Frankfurt plane.
The following dimensions; stature, waist height, shoulder-grip length, lower-arm length
and shoulder breadth; served as predictors because they are easy-to-measure and were obtained
thus:
Stature (ST): Measured as vertical distance from floor to crown of head in standing position
while the subject looks straight ahead.
Waist Height, Standing (WH): Measured as vertical distance from floor to the highest point on
the waist in standing position.
Shoulder-grip Length (SL): Measured as horizontal distance from the shoulder to the tip of the
longest finger in standing position.

Lower-arm length (LL): Measured as the horizontal distance from the elbow, when flexed at
900, to the tip of the longest finger in sitting position.
Shoulder breadth (SB): Measured as the maximum horizontal distance across the shoulder in
sitting position.
Anthropometric measurements that are needed for furniture design served as responses
because they are not easy-to-measure as the students have to sit in not too comfortable position.
These measurements are knee height, elbow rest height, popliteal height, shoulder height,
buttock-popliteal length and hip width (Agha, 2010; Agha and Alnahhal, 2012; Chung and
Wong, 2007; Gouvali and Boudolos, 2006; Knight and Noyes, 1999; Panagiotopoulou, et al.,
2004; Parcells, et al., 1999) and were obtained thus:
Knee Height (KH): Measured as vertical distance from the floor or the footrest to the uppermost
point on the knee in sitting position with knee flexed at 900.
Elbow Rest Height (EH): Measured as vertical distance from the sitting surface to the bottom of
the right elbow while the elbow was flexed at 900 and shoulder was flexed at 00.
Popliteal Height (P): Measured as the vertical distance from the floor or the footrest to the
underside of the thigh immediately behind the knee in the sitting position with knee flexed at 900.
Shoulder Height (SHH): Measured as the vertical distance from the sitting surface to the top of
the shoulder at the acromion position.
Buttock-Popliteal Length (BPL): Measured as the horizontal distance from the rear surface of
the buttock to the internal surface of the knee, or popliteal surface, with the knee flexed at 900.
Hip Width (HW): Measured as maximum horizontal distance across the hips in the sitting
position.

The measurements were carried out by two teams, each consisting of two people
(technicians trained in anthropometry). One team measured the students from Ogbomoso north
local government while the other team measured the students from Ogbomoso south local
government. The measurements were taken thrice and the average values were recorded as the
values of the respective dimensions. To ensure that intra-observer variations did not exist, 20
students (10 male, 10 female) were measured again by each team after two weeks. All the
measurements were statistically compared using paired t-test, but the variations were found to be
statistically insignificant.
2.5 Exhaustive search
Exhaustive search, implemented in ANFIS, is a very general problem-solving technique
that consists of systematically enumerating all possible candidates for the solution and checking
whether each candidate satisfies the problem's statement. It is an intensive approach and a
combinatorial function which selects the required number of inputs combination to be tried
during the search. Exhaustive search was used in this work to perform a thorough search of input
variables that best related with the output variables which gave the least difference between
training and checking error. That is, exhaustive search was employed to select the combination
of input variables that predicted each response with the least difference of root mean square error
(RMSE) between training and checking errors.
RMSE is defined as:
1
RMSE= √ ∑(obs-pre)2…………………………………………………………… (1)
n
Where n= sample size.
Obs= observed value(s).
Pre= predicted value(s).

The selected inputs were analyzed using response surface methodology, implemented in
design expert version 6.0.8. Descriptive statistics were reported (mean, minimum, maximum and
5th, 50th and 95th percentiles) to describe the anthropometric dimensions of subjects.
Second order polynomial response surface model, Equation (1), was fitted to each of the
response variable.
Equation (1)
Data were modeled by multiple regression analysis and the statistical significance of the
terms was examined by analysis of variance for each response. The statistical analysis of the data
and three dimensional plotting were performed using Design Expert Software (Stat-Ease 2002).
The adequacy of regression model was checked by lack-of ﬁt test, R 2, AdjR2, Pre R2, Adeq
Precision and F-test (Montgomery, 2001). The significance of F value was judged at 95%
confidence level. The regression coefficients were then used to make statistical calculation to
generate three-dimensional plots from the regression model.
3. Result
(Tables 1-4) presented some examples of independent variables selected by exhaustive
search for some responses. The selection is done on the basis of inputs combination that has the
least difference between training and checking errors.
3.1 Result of exhaustive search for male
From (Table 1), the inputs combination that has significant effect on the output (P) is ST
and WH because it has the training error of 1.0125, checking error of 1.1331 and the difference
of 0.1206 which is the least.

Table 1. . Inputs selection by exhaustive search for male.
The process was repeated for the entire responses and (Table 2) summarized the outcome
of the search.
Table 2. Summary of inputs selection by exhaustive search for male.
3.2 Result of exhaustive search for female
From (Table 3), the inputs combination that has significant effect on the output (SHH) is
ST and WH because it has the training error of 1.8559, checking error of 3.5274 and difference
of 1.6715 which is the least.
Table 3 . Inputs selection by exhaustive search for female.
The process was repeated for the entire responses and (Table 4) summarized the outcome
of the search.
Table 4 Summary of inputs selection by exhaustive search female.
It is well known that the way of presenting a study results may or may not make the
results of the research applicable (Mokdad and Ansari, 2009). For this reason, the anthropometric
measurements of students in this research are presented (Tables 5-6) in percentiles for ease of
use by furniture manufacturers.
Table 5 The anthropometric dimensions (cm); mean, minimum, maximum and 5th, 50th and 95th
percentiles; for male.
Table 6 The anthropometric dimensions (cm); mean, minimum, maximum and 5th, 50th and 95th
percentiles; for female.
3.3 Models presentation and analysis for male
Response 1: knee height

Two factors interaction (2FI) model is suggested by the design program for this response
to test for its adequacy and to describe its variation with independent variables. From anova test
in (Table 7), the Model F-value of 300.46 implies the model is significant. There is only a
0.01% chance that a "Model F-Value" this large could occur due to noise.
Table 7. anova test for KH
Values of "Prob > F" less than 0.0500 indicate model terms are significant. In this case A,
B and AB are significant model terms. Values greater than 0.1000 indicate the model terms are
not significant. The non-appearance of "Lack of Fit F-value" implies that the model perfectly
(100%) fit relative to the pure error.
Table 8. Post anova statistics for KH
From (Table 8), the "Pred R-Squared" of 0.9154 is in reasonable agreement with the "Adj
R-Squared" of 0.9192. "Adeq Precision" measures the signal to noise ratio. A ratio greater than
4 is desirable. The ratio of 67.051 indicates an adequate signal. This model can be used to
navigate the design space (Montgomery, 2001). In the same manner, other responses were
analyzed and the resulted is presented in (Table 9).
Table 9. Design summary for male
3.3.1 Model equations for male
Model equations are given in terms of coded factors and actual factors. Coded factors
indicate when the minimum and maximum values of the factors are represented by -1 and +1
respectively instead of their actual values.
The model in terms of coded factors is given by:
Equation (2)

The model in terms of actual factors is given by:
Equation (3)
Response 2: popliteal height
Equation (4)
Equation (5)
Response 3: shoulder height
Equation (6)
Equation (7)
Response 4: buttock popliteal length
Equation (8)
Equation (9)
Response 5: elbow height
Equation (10)
Equation (11)
Response 6: hip width

Equation (12)
Equation (13)
3.3.2 Diagnostic test-normal plots of residuals and predicted vs actual plots for male
Normal plots of residuals and predicted vs actual Plots do show how precisely the
responses are modeled. If all the points line up nicely and the deviation of points of the responses
from normality is insignificant, then the model is a very good one.
Take EH for example, from (Figures 1-2) below, it is clearly observed that the developed
models are very good models.
Figure 1. Normal plot of residuals for EH
Figure 2. Predicted vs actual for EH
3.3.3 Response surface plots analysis for male
To aid visualization of variation in responses with respect to independent variables, series
of three dimensional response surfaces were drawn using design expert software (Stat-Ease
2002).
Figure 3. Response surface plots for KH
(Figure 3) showed the interactive effect of ST and WH on the response (KH). At the low
WH value of 80.00, KH increases slightly from 45.41 to 46.60, as ST increases from 139.00 to
144.30. Furthermore, KH increases from 46.60 to 55.68, as ST increases from 144.30 to 178.80.
This implied that KH increases as ST increases. The relationship is almost a linear one. Also, at
high WH value of 104.30, KH increases from 50.53 to 56.70, as ST increases from 158.00 to

174.00. Furthermore, KH increases slightly from 56.70 to 57.14, as ST increases from 174.00 to
178.80. This implied that KH increases as ST increases in almost a linear relationship.
Figure 4. Response surface plots for P
(Figure 4) showed the interactive effect of ST and WH on the response P. At the low WH
value of 8.00, P increases slightly from 35.44 to 36.46, as ST increases from 139.00 to 144.30.
Furthermore, P increases from 36.46 to 45.02, as ST increases from 144.30 to 178.80. This
implied that P increases as ST increases. Also, at high WH value of 104.30, P increases from
42.74 to 45.43, as ST increases from 139.00 to 174.00. Furthermore, P increases slightly
form45.43 to 45.73, as ST increases from 174.00 to 178.80. This implied that P increases slightly
as ST increases.
Figure 5. Response surface plots for SHH
(Figure 5) showed the interactive effect of ST and WH on the response (SHH). At the
low WH value of 80.00, SHH increases from 42.69 to 44.41, as ST increases from 139.00 to
144.30. Furthermore, SHH increases from 44.41 to 55.44, as ST increases from 144.30 to
178.80. This implied that SHH increases linearly as ST increases. Also, at high SB value of
33.90, SHH increases from 32.89 to 52.59, as ST increases from 134.00 to 174.00. Furthermore,
SHH increases from 52.59 to 54.27, as ST increases from 174.00 to 178.80. This implied that P
increases linearly as ST increases.
Figure 6. Response surface plots for BPL

(Figure 6) showed the interactive effect of ST and WH on the response (BPL). At the low
WH value of 80.00, BPL increases from 41.73 to 42.68, as ST increases from 139.00 to 144.30.
Furthermore, BPL decreases from 42.68 to 42.02, as ST increases from 144.30 to 178.80. This
implied that BPL increases slightly as ST increases and later decreases slightly as ST increases.
Also, at high WH value of 104.30, BPL increases from 40.28 to 51.03, as ST increases from
139.00 to 174.00. Furthermore, BPL increases slightly from 51.03 to 51.07, as ST increases from
174.00 to 178.80.
Figure 7. Response surface plots for EH
(Figure 7) showed the interactive effect of SB and WH on the response (EH). At the low
WH value of 80.00, EH increases from 14.09 to 17.35, as SB increases from 21.60 to 26.70.
Furthermore, EH increases from 17.35 to 21.39, as SB increases from 26.70 to 33.30. This
implied that EH increases as SB increases. Also, at high WH value of 104.30, EH decreases
slightly from 16.99 to 15.50, as SB increases from 21.60 to 29.90. Furthermore, EH decreases
slightly from 15.50 to 14.67, as SB increases from 29.90 to 33.30. This implied that EH
decreases slightly as SB increases.
Figure 8. Response surface plots for HW
(Figure 8) showed the interactive effect of SB and WH on the response (HW). At the low
WH value of 80.00, HW increases from 24.70 to 27.25, as SB increases from 21.60 to 26.70.
Furthermore, HW increases from 27.25 to 30.53, as SB increases from 26.70 to 33.30. This
implied that HW increases as SB increases. Also, at high WH value of 104.30, HW increases
from 26.65 to 30.51 as SB increases from 21.60 to 29.90. Furthermore, HW increases slightly

from 30.51 to 30.53, as SB increases from 29.90 to 31.72. This implied that HW increases as SB
increases and later increases slightly as SB increases.
3.4 Models presentation and analysis for female
Quadratic model is suggested by the design program for this response to test for its
adequacy and to describe its variation with independent variables. From anova test in (Table 10),
the Model F-value of 4267.87 implies the model is significant. There is only a 0.01% chance
that a "Model F-Value" this large could occur due to noise.
Table 10. anova test for EH
Values of "Prob > F" less than 0.0500 indicate model terms are significant. In this case A,
B, A2 and AB are significant model terms. Values greater than 0.1000 indicate the model terms
are not significant. The "Lack of Fit F-value" of 3.96 implies that there is 9.27% chance that a
"Lack of Fit F-value" this large could occur due to noise.
Table 11. Post anova statistics for EH
From (Table 11), the "Pred R-Squared" of 0.9952 is in reasonable agreement with the
"Adj R-Squared" of 0.9963. "Adeq Precision" measures the signal to noise ratio. A ratio greater
than 4 is desirable. The ratio of 294.750 indicates an adequate signal. This model can be used to
navigate the design space [18]. In the same manner, other responses were analyzed and the
resulted is presented in (Table 12).
Table 12. Design summary for female

3.4.1 Model equations for female
Model equations are given in terms of coded factors and actual factors. Coded factors
indicate when the minimum and maximum values of the factors are represented by -1 and +1
respectively instead of their actual values.
Equation (14)
Equation (15)
Equation (16)
Equation (17)
Equation (18)
Equation (19)
Equation (20)

Equation (21)
Equation (22)
Equation (23)
Equation (24)
Equation (25)
3.4.2 Diagnostic test-normal plots of residuals and predicted vs actual plots for female
Normal plots of residuals and predicted vs actual Plots do show how precisely the
responses are modeled. If all the points line up nicely and the deviation of points of the responses
from normality is insignificant, then the model is a very good one.
Take BPL for example, from (Figures 9-10) below, it is clearly seen that the developed
models are very good models.
Figure 9. Normal plot of residuals for BPL
Figure 10. Predicted vs actual for BPL
3.4.3 Response surface plots analysis for female
To aid visualization of variation in responses with respect to independent variables, series
of three dimensional response surfaces were drawn using Design Expert Software (Stat-Ease
2002).

Figure 11. Response surface plots for EH
(Figure 11) showed the interactive effect of ST and WH on the response (EH). At the low
WH value of 83.00, EH increases slightly from 15.54 to 16.40, as ST increases from 145.50 to
148.60. Furthermore, EH increases from 16.40 to 21.76, as ST increases from 148.60 to 169.00.
This implied that KH increases as ST increases. The relationship is almost a linear one. Also, at
high WH value of 106.00, EH decreases slightly from 17.60 to 16.81, as ST increases from
145.50 to 164.40. Furthermore, EH further decreases slightly from 16.81 to 16.78, as ST
increases from 164.40 to 169.00. This implied that EH decreases slightly as ST increases in
almost a linear relationship.
Figure 12. Response surface plots for SHH
(Figure 12) showed the interactive effect of ST and WH on the response (SHH). At the
low WH value of 83.00, SHH increases slightly from 46.31 to 47.59, as ST increases from
145.50 to 148.60. Furthermore, SHH increases from 47.59 to 54.81, as ST increases from 148.60
to 169.00. This implied that SHH increases as ST increases. Also, at high WH value of 106.00,
SHH increases from 45.20 to 51.89, as ST increases from 145.50 to 164.40. Furthermore, SHH
increases slightly from 51.89 to 53.73, as ST increases from 164.40 to 169.00. This implied that
SHH increases slightly as ST increases.
Figure 13. Response surface plots for KH
(Figure 13) showed the interactive effect of SB and WH on the response (KH). At the
low WH value of 80.00, KH increases slightly from 47.72 to 48.18, as SB increases from 22.70

to 27.90. Furthermore, KH increases slightly from 48.18 to 48.78, as SB increases from 27.90 to
32.60. This implied that KH increases slightly and linearly as ST increases. Also, at high WH
value of 106.00, KH increases slightly from 53.35 to 54.84, as SB increases from 22.70 to 31.20.
Furthermore, KH increases slightly from 54.84 to 55.03, as SB also increases slightly from 31.20
to 32.60. This implied that KH increases slightly and linearly as ST increases.
Figure 14. Response surface plots for P
(Figure 14) showed the interactive effect of SB and WH on the response (P). At the low
WH value of 83.00, P decreases slightly from 38.11 to 38.07, as SB increases from 22.70 to
27.90. Furthermore, P further decreases slightly from 38.07 to 37.98, as SB increases from 27.90
to 32.60. This implied that P decreases slightly as SB increases. Also, at high WH value of
106.00, P slightly increases from 44.05 to 44.25, as SB increases from 22.70 to 31.20.
Furthermore, the value of P remains constant, 44.25, as SB increases from 31.20 to 32.60.
Figure 15. Response surface plots for BPL
(Figure 15) showed the interactive effect of SB and WH on the response (BPL). At the
low WH value of 83.00, BPL increases from 43.46 to 45.01, as SB increases from 22.70 to
27.90. Furthermore, BPL increases from slightly 45.01 to 46.55, as SB increases from 27.90 to
32.60. This implied that BPL increases as SB increases. Also, at high WH value of 106.00, BPL
increases slightly from 51.57 to 53.90, as SB increases from 22.70 to 31.20. Furthermore, BPL
increases slightly from 53.90 to 54.38, as SB increases slightly from 31.20 to 32.60. This implied
that BPL increases slightly as SB increases.

Figure 16. Response surface plots for HW
(Figure 16) showed the interactive effect of SB and WH on the response (HW). At the
low WH value of 83.00, HW increases from 27.89 to 30.49, as SB increases from 22.70 to 27.90.
Furthermore, HW increases from 30.49 to 37.70, as SB increases from 27.90 to 32.60. This
implied that HW increases as SB increases. Also, at high WH value of 106.00, HW decreases
from 36.54 to 31.73 as SB increases from 22.70 to 31.20. Furthermore, HW increases slightly
from 31.73 to 32.12, as SB also increases slightly from 31.20 to 32.60. This implied that HW
decreases as SB increases and later increases slightly as SB increases slightly.
(Table 13) gives the summary of co-efficient of determination and co-efficient of
variation of the middle classes.
Table 13. Co-efficient of determination (R2) and co-efficient of variation (C.V.) of all the
responses.
4. Discussions.
4.1 Efficiencies of the models.
Co-efficient of determination (R2) and co-efficient of variation (C.V) are the commonly
used performance measures for detecting the efficiency of predictive models (Agha and
Alnahhal, 2012). High value of R2 and low value of C.V are desirable. In this study, twelve
models were developed and the adjusted co-efficient of determination (R 2) is greater than 0.90 in
all. In general, all the models showed good predictive ability (efficiency) as can be observed in
(Table 13). In fact, 83.33% of the models have adjusted R2 value of over 95%. This confirmed
the validity of the models.
Liyana-Pathirana and Shahidi (2005) stated that a high coefficient of variation (CV)
demonstrates that variation in the mean value is large and does not sufficiently generate an

acceptable response model. Therefore, CV < 10% has been suggested suitable in predicting the
response surface models. From table 6, CV <= 1.43% for all the models. Thus, the models
exhibited high predictive ability (they are efficient).
4.2 Discussions of the models.
Using ergonomic principles, the current study developed 12 models considered necessary
for the design of furniture for middle class students. While the relationships among standing
height and length dimensions have usually been assumed linear, the present study showed
otherwise. Out of the 12 models developed, 11 of them have non-linear relationships
(representing 91.67%). Moreover, the present study obtained a higher value of R 2 which is
0.9725 and 0.9961 compared with R2_ 0.81 (r _ 0.90) obtained by Castellucci et al (2010) and
0.844 obtained by Ismaila et al., (2014) for the model for predicting buttock popliteal length.
Also, the value of R2 for EH in the present study is 0.9774 and 0.9935 which is far higher than
0.416 obtained by Ismaila et al., (2014). Agha and Alnahhal (2012) reported that EH cannot be
predicted but rather measured. However, the present study showed otherwise. (Table 14)
presented the performance of the models developed in this study compared with the previous
ones.
Table 14. Performance comparism of the developed models with those of previous researches.
Ismaila et al., (2014) stated that it can be very expensive in developing countries to obtain
anthropometric data when needed, and as such, measuring one anthropometric value to
determine others would be helpful and affordable. Although economic reason is important but, at
the same time, adequacy and effectiveness of the predictive models cannot be compromised. The
current study took these three factors (economic reason, adequacy and effectiveness) into

consideration. Using three anthropometric dimensions to predict six needed for the design of
school furniture is justifiable in view of the high predictive ability of the models.
5. Conclusion.
Ergonomic design of products and workplaces demands up-to-date anthropometric data
which, many times, are not handy. The present study, therefore, proposed 12 models that can be
used to estimate various anthropometric dimensions necessary for the design of furniture for use
of middle class students in Ogbomoso, South Western Nigeria. The furniture industry would
benefit from these economical, adequate and effective predictive models.
Disclosure Statement: There is no financial interest or benefit arising from the direct
applications of this research.
About the authors
S.O. Oladapo, is a PhD student in industrial engineering in the Department of Industrial and Production
Engineering, University of Ibadan. His research interests include Ergonomics, Safety and Project Management.
O.G. Akanbi, PhD, is a senior lecturer in industrial engineering in the Department of Industrial and Production
Engineering, University of Ibadan. His research interests include Ergonomics, Safety, Operations Research and
Systems Engineering. He is a Registered Engineer with COREN and a Member of Nigerian Society of Engineers.
References
(1) Agha, S.R. 2010. “School furniture match to students’ anthropometry in Gaza Strip.”
Ergonomics 53 (3): 344-354.
(2) Agha, S.R., and M.J. Alnahhal. 2012. “Neural network and multiple linear regressions to
predict school children dimensions for Ergonomic school furniture design.” Applied Ergonomics
43: 979-984.

(3) Barli, O., M.R./Sari, Elmali, D. and E. Aydintan. 2006. “Anthropometric evaluation of the
Creches children furniture in Turkey.” Coll. Antropol. 30 (4): 853-865.
(4) Castellucci, H.I., P.M. Arezes, and C.A. Viviani. 2010. “Mismatch between classroom
furniture and anthropometric measures in Chilean schools.” Applied Ergonomics 41 (4): 563-
568.
(5) Chung, J.W.Y., and T.K.S. Wong. 2007. “Anthropometric evaluation for primary school
furniture design.” Ergonomics 50 (3): 323-334.
(6) Corlett, E.N. 2006. “Background to sitting at work: research-based requirements for the
design of work seats.” Ergonomics 49:1538-1546.
(7) Dianat, I., M.A. Karimi, A.A., Hashemi, and Bahrampour, S. 2013. “Classroom furniture and
anthropometric data.” Applied Ergonomics 44 (1): 101-108.
(8) Garcia-Acosta, G., and K. Lange-Morales. 2007. “Definition of sizes for the design of school
furniture for Bogota schools based on anthropometric criteria.” Ergonomics 50 (10): 1626-1642.
(9) Gouvali, M.K., and K. Boudolos. 2006. “Match between school furniture dimensions and
children anthropometry.” Applied Ergonomics 37 (6): 765-773.
(10) Grimmer, K., and M. Williams. 2000. “Gender-age environment associates of adolescent
low back pain.” Applied Ergonomics 31: 343-360.
(11) High, R. 2000. “Important factors in designing statistical power analysis studies.” Accessed
July 28 2014.
http://cc.uoregon.edu/cnews/statpower.html.
(12) Ismaila, S.O., O.G. Akanbi, and C.N. Ngassa. 2014. “Models for estimating the
anthropometric dimensions using standing height for furniture design.” Journal of engineering,
design and technology 12 (3): 336-347.

(13) Jeong, B.Y., and K.S.S Park. 1990. “Sex difference in anthropometry for school furniture
design.” Ergonomics 33 (12): 1511-1521.
(14) Knight, G., and J. Noyes. 1999. “Children’s behaviour and the design of school furniture.”
Ergonomics 42 (5): 747-760.
(15) Liyana-Pathirana, C., and F. Shahidi. 2005. “Optimization of extraction of phenolic
compounds from wheat using response surface methodology.” Food Chem. 93:47-56.
(16) Mokdad, M., and M. AI- Ansari. 2009. “Anthropometrics for the design of Bahraini school
furniture.”. International Journal of Industrial Ergonomics 39: 728-735.
(17) Molenbroek, J.F.M., and Y.M.T. Ramaekers. 1996. “Anthropometric design of a size
system for school furniture.” In Robertson, S.A. (Ed.). Proceedings of the annual conference of
the Ergonomics society: Contemporary Ergonomics, 130-135, Taylor and Francis:, London.
(18) Montgomery, D.C. 2001. Design and analysis of experiments (5th ed.). Wiley, New York
(19) Niekerk, S., Q.A. Louw, Grimmer-Somers, A., and J. Harvey. 2013. “The anthropometric
match between high school learners of the caps metropolitan area, Western cape, South Africa
and their computer Workstation at school.” Applied Ergonomics 44 (3): 366-371.
(20) Oyewole, S.A., J.M. Haight, and A. Freivalds. 2010. “The ergonomic design of classroom
furniture/computer work station for first graders in the elementary school.” International Journal
of Industrial Ergonomics 40 (4): 437-447.
(21) Panagiotopoulou, G., K. Christoulas, Papanckolaou, A., and Mandroukas, K. 2004.
“Classroom furniture dimensions and anthropometric measures in primary school.” Applied
Ergonomics 35 (2): 121-128.

(22) Parcells, C., S. Manfred, and Hubbard, R. 1999. “Mismatch of classroom furniture and body
dimensions, Empirical findings and health implications.” Journal of Adolescent Health 24 (4):
265-273.
(23) Saarni, L., C-H. Nygard, Kaukiainen, A., and Rimpela, A. 2007. “Are the desks and chairs
at school appropriate?.” Ergonomics 50 (10): 1561 – 1570.
(24) Savanur, C.S., C.R. Altekar, and De, A. 2007. “Lack of conformity between Indian
classroom furniture and student dimensions: proposed future seat/table dimensions.” Ergonomics
50 (10): 1612-1625.
(25) Straker, L., B. Maslen, R. Burgress-Limerick, P. Johnson, and Dennerlein, J. 2010.
“Evidence-based guideline for wise use of computers by children: physical development
guidelines.” Ergonomics 53 (4): 458-477.
(26) Turnay, M., and K. Melemez. 2008. “Analysis of biomechanical and anthropometric
parameters on classroom furniture design.” African Journal of Biotechnology 7 (8): 1081-1086.

Models For Predicting Anthropometric Dimensions of Students Needed For Ergonomic School Furniture Design

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Models For Predicting Anthropometric Dimensions of Students Needed For Ergonomic School Furniture Design

Uploaded by

Copyright:

Available Formats

Models for predicting anthropometric dimensions of students needed for ergonomic

school furniture design

S.O. Oladapoa; O.G. Akanbib.

University of Ibadan, Ibadan, Nigeria.

University of Ibadan, Ibadan, Nigeria.

S.O. Oladapo is the corresponding author and can be contacted at:

dapobola@gmail.com, phone number: (+234)08060530526

*Corresponding author. Email: dapobola@gmail.com

reduced if not eliminated.

Key words: furniture, students, dimensions, models, school, design.

facilitate learning, monitoring of behavior and performance and to minimize distracting

interactions (Knight and Noyes, 1999).

tiredness, psychological manifestations and musculoskeletal dysfunction can increase during

*Corresponding author. Email: dapobola@gmail.com

Parcells, et al., 1999)..

consuming and energy-sharping anthropometric dimensions needed for ergonomic school

2. Material and method.

2.1 Study design.

*Corresponding author. Email: dapobola@gmail.com

representative sample of secondary school students in the area.

2.2 Participant Sample.

furniture was also reported by Chung and Wong (2007).

2.3 Statistical power analysis for the present study

*Corresponding author. Email: dapobola@gmail.com

difference meant more power for the research.

2.4 Description of anthropometric measurements

the subjects were wearing light clothing and were barefooted.

were in good contact and their trunk was upright.

while the subject looks straight ahead.

the waist in standing position.

longest finger in standing position.

*Corresponding author. Email: dapobola@gmail.com

900, to the tip of the longest finger in sitting position.

2004; Parcells, et al., 1999) and were obtained thus:

point on the knee in sitting position with knee flexed at 900.

the shoulder at the acromion position.

*Corresponding author. Email: dapobola@gmail.com

2.5 Exhaustive search

Exhaustive search, implemented in ANFIS, is a very general problem-solving technique

(RMSE) between training and checking errors.

RMSE is defined as:

Where n= sample size.

Obs= observed value(s).

Pre= predicted value(s).

*Corresponding author. Email: dapobola@gmail.com

generate three-dimensional plots from the regression model.

least difference between training and checking errors.

3.1 Result of exhaustive search for male

of 0.1206 which is the least.

*Corresponding author. Email: dapobola@gmail.com

Table 2. Summary of inputs selection by exhaustive search for male.

3.2 Result of exhaustive search for female

of 1.6715 which is the least.

Table 3 . Inputs selection by exhaustive search for female.

Table 4 Summary of inputs selection by exhaustive search female.

use by furniture manufacturers.

percentiles; for male.

percentiles; for female.

3.3 Models presentation and analysis for male

Response 1: knee height

*Corresponding author. Email: dapobola@gmail.com

(100%) fit relative to the pure error.

Table 8. Post anova statistics for KH