J. AMER. SOC. HORT. SCI. 121(5):927–936. 1996.

Penetrometer Measurement of Apple and Kiwifruit

Firmness: Operator and Instrument Differences
F.R. Harker1, J.H. Maindonald, and P.J. Jackson
Horticulture and Food Research Institute of New Zealand, Mt. Albert Research Centre, Private Bag 92
169, Auckland, New Zealand
Additional index words. Instron, puncture, twist test, Malus domestica, Actinidia deliciosa
Abstract. Flesh firmness is a characteristic used to indicate fruit quality. Experimental design and data analysis are
important when comparing devices that measure fruit firmness. We compared the Effegi penetrometer operated by hand,
mounted in a drill press and then operated by hand, and mounted on a motorized drive and operated remotely; the hand-
operated EPT pressure tester; the Instron with an Effegi probe; and a hand-operated prototype of the twist tester. Devices
varied in operator differences and precision. Comparisons between devices were at the within-fruit level of variability and,
therefore, more precise than comparisons where different device-operators used different fruit. We demonstrate
statistical methods that are appropriate for making the comparisons of interest and discuss the possible cause of
differences between operators and between devices. We also discuss how the mechanical properties of the devices may
affect results and consider implications for their practical use. In this study, we found the precision of discrimination
between soft and hard apples was best using the Instron in 1992, while the Instron and hand-held Effegi penetrometer were
comparable in 1991. For kiwifruit, the hand-held Effegi penetrometer consistently gave the most precise measurements
of softening in 1991, while the twist test was the most precise in 1992.

Fruit quality assessments usually involve measuring flesh firm- tween soft and hard fruit. The usefulness of the design was
ness using a penetrometer. Generally, a probe, with either a flat or evaluated by assessing the factor by which the number of measure-
convex tip, is driven into the flesh, and the maximum force is ments would need to be increased to achieve similar precision with
recorded. Early hand-held penetrometers were developed by a less controlled design.
Magness and Taylor (1925), and from these devices, a range of Further design gains are possible. With large fruit, such as
devices have been developed by different companies (reviewed by apples, as many as eight measurements per fruit may be possible
Bourne, 1982). A number of studies have examined the mechanics without interference between results (Abbott et al., 1976). These
of puncture testing (Voisey, 1977, and references therein) and multiple measurements would allow one pair of measurements for
recommendations on using penetrometers include those of Blanpied each of four device-operator combinations on each fruit. Biases
et al. (1978), Smith (1985), and Watkins and Harman (1981). that affect a particular device can, in principle, be corrected by a
The wide use of firmness measurements has led to the periodic suitable calibration. Inherent lack of precision can be cured only by
development of new devices for measuring firmness (e.g., Studman redesign of the device.
and Yuwana, 1992). Any adequate evaluation of their performance
relative to existing equipment must consider operator differences, Materials and Methods
precision, and ability to detect texture change, as well as differ-
ences in measurement of absolute firmness. Effective experimen- Plant material. In the 1991 and 1992 commercial apple har-
tal design is critical for comparing devices and comparing different vests, fruit (Malus domestica L. cv. Granny Smith) harvested in
ways of using the devices. Hawkes Bay were collected from the New Zealand Apple and Pear
Previous studies have compared different types of penetrom- Marketing Board depot in Hastings and were transported to
eters (e.g., Abbott et al., 1976; Bongers, 1992) and investigated Auckland. Kiwifruit [Actinidia deliciosa (A. Chev) C.F. Liang et
how experimental design (replicate number and sample size) A.R. Ferguson cv. Hayward] were obtained from the HortResearch
affects detection of firmness differences (Saltveit, 1978; Orchard (Kumeu, Auckland) in both years. Kiwifruit and apples
Worthington and Yeatman, 1968). In our study, we used a bal- were placed in separate cool stores at 0 °C, and firmness experi-
anced, incomplete blocks (paired comparison) experimental de- ments were undertaken within the first week and after storage.
sign together with modern statistical analysis approaches to compare Apples were stored for 24 (1991) or 18 weeks (1992), and kiwifruit
different methods of measuring firmness. For relatively modest were stored for 17 (1991) or 15 (1992) weeks. All fruit were held
effort, our approach has provided information on differences at 20 °C for 1 d before firmness was measured.
associated with different operators and different methods and Penetrometer devices. Measurements made in this study in-
information on precision with which methods discriminate be- volve driving an Effegi probe with a convex tip into whole fruit.
However, the mechanism used to drive the probe into the fruit
differed. Probes with diameters of 11.1 and 7.9 mm were used to
Received for publication 2 Jan. 1996. Accepted for publication 13 May 1996. We
acknowledge the help of B.A. Cregoe, H.J. Elgar, C.B. Watkins, and A. White who
measure firmness of apples and kiwifruit, respectively. These
generously agreed to participate in this study. Mention of a trademark or propriety probe sizes reflect standard methods for the respective fruit types.
does not constitute a guarantee or warranty of the product by the Horticulture and The Effegi hand-held penetrometer (Facchini, Alfonsine, Italy)
Food Research Institute of New Zealand and does not imply its approval to the was the simplest of the devices (Fig. 1a). The fruit was held steady
exclusion of other products that also may be suitable. The cost of publishing this on a firm surface and the probe was pushed into the fruit to a depth
paper was defrayed in part by the payment of page charges. Under postal regula-
tions, this paper therefore must be hereby marked advertisement solely to indicate of 8 mm, corresponding to a mark inscribed on the shaft of the
this fact. probe (Watkins and Harman, 1981). In addition, the Effegi pen-
1
To whom reprint requests should be addressed. etrometer was mounted either in a drill press (Black and Decker,

J. AMER. SOC. HORT. SCI. 121(5):927–936. 1996. 927

Fig. 1. Devices used to measure fruit firmness: (A) Effegi penetrometer, (B) EPT pressure tester, (C) Effegi penetrometer mounted in a motorized drive, and (D) the twist
test (with insert showing blade). Instron and Effegi penetrometer mounted in drill press are not shown.

928 J. AMER. SOC. HORT. SCI. 121(5):927–936. 1996.

Hampstead, Md.) or a motorized press (Fig. 1c). The motorized design was a balanced incomplete block design, with two devices
press was developed by the Engineering Development Group of per block (fruit) (Gacula and Singh, 1984; Pearce, 1983). There
HortResearch as a standardized method of measuring kiwifruit were two results for each device in each block. A single fruit
firmness. The horizontal beam was driven downward at 1500 constituted a “block.”
mm·min–1. The end of the test was detected electronically when the Statistical analysis. From examining plots of residuals against
probe had penetrated 8 mm into the fruit. The beam then returned fitted values in the analysis of variance (ANOVA) calculations
automatically to its original position. described later in this paper, we concluded that using log (firm-
The EPT pressure tester (Lake City Technical Products, ness) would best ensure that variances were comparable over the
Kelowna, Canada; Fig. 1b) also is hand-operated. The penetrom- whole range of firmness. To reflect the experimental design
eter probe is located at the end of a pivoting arm, and the arm is correctly, we needed an ANOVA table (Table 2) in which there
pulled down until the probe enters the fruit to a depth of 8 mm. A were three major strata, or levels, of variability. These were
spring returns the arm to its original position, and a load cell, between fruit (“fruit” stratum), between different pairs of measure-
located on a leaf spring within the arm, measures the force. The ments on the same fruit (“fruit × device” stratum), and within pairs
output is digital and can be imported into a computer or printed of measurements made by one device–operator combination (“fruit
directly. The pressure tester interprets information on the shape of × device × units” stratum). Relatively imprecise information on the
the penetration curve and provides messages when the measure- comparison between device–operator combinations appears in the
ment was taken “too fast” or “too slow”, as well as the prompts “no interfruit (or interblock) analysis of the fruit stratum. More precise
slope” and “bad sample” when apples are overly soft or have inner information on this comparison appears in the intrafruit (or
pulpiness or bruises. The EPT pressure tester was used in the “MT intrablock) analysis of the fruit × device stratum. For a discussion
off” mode. of the interblock and intrablock analysis for a balanced incomplete
Materials testing machines are widely used in assessing texture design, see for example Gacula and Singh (1984). The final or fruit
in foods (Bourne, 1982). We used an Instron (model 4301) mate- × device × units stratum contains no information on the compari-
rials testing machine (Instron, Canton, Mass.) to apply penetrom- son between device–operator combinations. The sum of squares in
eter tests to fruit. The speed of penetration was set at 240 mm· this stratum (180 × 0.0085) is the total of the sums of squares about
min –1, and the test was stopped after penetration to 8 mm deep. the mean for the two results for each device–operator combination
The twist tester consists of a blade fixed at 90° to a horizontal used in each of the 90 fruit.
spindle (Studman and Yuwana, 1992) (Fig. 1d). The fruit is pushed For our analysis, we used the Genstat ANOVA routine
onto the spindle and rotated around the spindle until the blade
crushes the flesh and the fruit turns freely. The resisting moment Table 1. An example showing the type of experimental design used to
is generated by a counterweight attached to an offset arm. The compare penetrometers. The table indicates how four operator–device
angle that the fruit rotated before tissue failure was measured using combinations were tested against each other. For each operator–device
a potentiometer connected to a chart recorder (model R-01; combination, firmness was measured twice, one on each of two
opposing sides of the fruit. In this example, there are six pairwise
Rikadenki, Tokyo). This angle then was converted into a crushing comparisons, against 45 and 65 pairwise comparisons in actual experi-
strength using formulas described by Studman and Yuwana (1992). ments in 1991 and 1992, respectively.
In 1991, we compared Effegi hand-held penetrometer, Effegi
penetrometer mounted in a motorized press, EPT pressure tester, Operator–device Operator–device combination
and Instron. In 1992, we compared Effegi hand-held penetrometer, combination Danz-Handy Dan-EPT Cloe-Hand Cloe-EPT
Effegi penetrometer mounted in a drill press, Instron, and twist Dan-hand 1x, 2 3, 4 5, 6
test. All measurements were made to the nearest 1 N, except for the Dan-EPT 7, 8 9, 10
twist test, which was measured to the nearest 7 KPa. Cloe-hand 11, 12
Operators. All operators were members of the postharvest Cloe-EPT
group of HortResearch and had at least 1 year of prior experience z Operators
are identified by pseudonyms.
using hand-held Effegi penetrometers. None of the operators had yInstruments
used the EPT pressure tester, twist tester, or a drill press to operate are identified by abbreviations: hand = hand-held Effegi
penetrometer and EPT = EPT pressure tester.
the Effegi penetrometer before these experiments. Operator were xEach number identifies an individual fruit.
allowed sufficient measurements to familiarize themselves with
the devices before the start of each series of measurements. Table 2. Overall analysis of variance table for stored kiwifruit in 1991. The
Pseudonyms identify operators. Genstat statement BLOCK fruit/device gives the error structure needed
Experimental design. Slices of skin were removed at four to get this table in which each analysis of variance mean square is
equidistant points around the equator of each fruit. Firmness then associated with its appropriate residual or “error” mean square. The
was measured on opposite sides of the fruit for each device– residual mean square is a variance estimate for variation in that
stratum.
operator combination before passing the fruit to the next device–
operator combination for firmness measurements on the remaining Source of Degrees of Mean
opposite sides. This process allowed us to compare the device– variation freedom square
operator combinations relative to variation between different Fruit stratum (interfruit analysis)
measurements on the same fruit, thus, excluding any fruit-to-fruit Device 9 2.233
component of variability. Each person (Amy, Cloe, Dan, and Ed in Residual 80 0.127
1991 and Ben, Dan, Ed, and Fay in 1992) used each instrument Fruit × device stratum (intrafruit analysis)
with the exception of the Instron and motorized press, which were Device–operator 9 3.146
presumed to be operator independent and, thus, were operated by Residual 81 0.0096
F.R.H. Fruit × device × units stratum 180 0.0085
Two fruit were used, giving two sets of measurements for each Total 359
pair of device–operator combinations (Table 1). Formally, the

J. AMER. SOC. HORT. SCI. 121(5):927–936. 1996. 929

(Maindonald, 1992; Payne et al., 1993; Pearce, 1983), which is mean square (variance estimate = sum of squares ÷ degrees of
able to give the table described in the previous paragraph. Table 2 freedom) based on differences between measurements on opposite
may be obtained from an ANOVA table that ignores the hierarchi- sides was used as a measure of inherent variability (lack of
cal error structure by a suitable repartitioning and grouping of the precision) of each device–operator combination. Boxplots were
entries in the rows. Typically, it is expected that the error mean used to display fruit-to-fruit variation and differences of individual
square for differences between fruit (i.e., in what Table 2 calls the measurements for a device–operator combination from the fruit
fruit stratum) will be larger than the error mean square for compari- mean. These helped to identify outliers. We used the robloc
sons between different devices on the same fruit (i.e., in the fruit function in S-PLUS (Statistical Sciences, 1991) to determine
× device stratum. By using a design and accompanying analysis robust estimates for the variance based on differences between
that allows comparison of devices relative to this second and measurements on opposite sides. Such estimates are not affected
smaller error mean square, we avoid the unnecessary “noise” that by an occasional large outlier.
differences between fruit would otherwise add to the comparison We used Bartlett’s test for homogeneity of variance (Miller,
of devices. This analysis may be extended to split the stratum 1986) to check on apparent variance differences between device–
variances into components of variance, as in the Genstat REML operator combinations. Because Bartlett’s test is sensitive to
analysis (Payne et al., 1993). We have not taken this further step as nonnormality, it was important to use this test with checks on
it would not assist data interpretation. outliers and with a low significance level (P = 0.01).
ANOVA tables, tables of means, and standard errors of differ- Analysis of twist tester measurements. The data presented in
ences (SEDs) of means were determined for each combination of Fig. 2 allowed us to relate kiwifruit twist test measurements (1992
year, before or after storage, and fruit (kiwifruit or apple). It was only) to penetrometer measurements. To compare variability be-
necessary to make comparisons between storage and harvest tween the two types of device, we needed to reduce them to a
results relative to between-fruit variation. common scale. Because we wished to place the two scales on an
To get an approximate overall comparison between device– equal footing, neither of the regressions log y on log x nor log x on
operator combinations, we made the simplifying assumption that log y was appropriate for this purpose. Instead, we estimated a
within-fruit variances for the different device–operator combina- functional relationship, as described in Kendall and Stuart (1979,
tions were similar. The overall SEDs and least significant differ- p. 409–412). We assumed that with y = twist test result and x =
ences that are presented should be used as broad indicators of penetrometer result, the errors in log y and log x have equal
differences. The variance ratios for testing for differences between variance. Then log y = 2.428 + 1.164 (SE = 0.025) log x. The line
device–operator combinations always were so large that formal is shown in Fig. 3. The same transformation has been used for apple
use of an F test was unnecessary. twist measurements to get comparability with other devices.
In addition, analyses were performed for each individual de- Over the range of measurements used, the relationship was
vice–operator combination. As before, these partitioned the total linear. The approach can be extended to handle situations where
sum of squares into a sum of squares for between-fruit variation the relationship is nonlinear. For example, one might precede the
and two sets of sums of squares for within-fruit variation. The calibration just described with using a smoothing function to

Fig. 2. Relationship between penetrometer and twist test measurements of kiwifruit firmness. Measurements obtained using the twist test are plotted against firmness
according to penetrometer measurements. The solid line is obtained as described in the text. Broken lines (······ and ----) are smoothed curves using S-PLUS loess
smoothing function for y on x and for x on y, respectively.

930 J. AMER. SOC. HORT. SCI. 121(5):927–936. 1996.

Fig. 3. Relationship between firmness of opposite and adjacent quarters of kiwifruit. Firmness was measured using an Instron with a probe 7.9 mm in diameter.
Measurements x1, x2, x3, and x4 were made in order around the fruit equator. The graphs show (A) sums of pairs of opposite sides plotted against each other and (B
and C) sums of pairs of adjacent sides plotted against each other. Values are in newtons.

provide a transformation to a Table 3. Comparison of between- and within-fruit (fruit × device stratum) mean squares. Notice that between-
linear relationship. The S-PLUS fruit mean squares generally are much larger than within-fruit mean squares. This difference has
“loess” smoothing function implications for experimental design. The mean squares have been used to calculate the estimates of the
(Chambers and Hastie, 1991) is number of fruit required to achieve the same precision when different device–operator combinations are
used with different fruit that appear in the last column.
suitable for this purpose. The
presentation of data in the fig- Mean squares Fruit no.
ures is based on user styles that z
Fruit Between Within Eff. No. Alt. expt.y
were discussed and recom-
1991
mended by Maindonald (1992).
Kiwifruit Harvest 0.40 (80)x 0.024 (81)x 0.556 90 90 × 9.3
Results and Discussion Storage 0.13 0.0096 0.556 90 90 × 7.3
Apple Harvest 0.0068 0.0026 0.556 90 90 × 1.5
Preliminary investigations. Storage 0.014 0.0021 0.556 90 90 × 3.7
There were four measurements 1992
on each fruit, two (on opposite Kiwifruit Harvest 0.15 (143)x 0.015 (144)x
0.542 156 156 × 5.6
sides) with one device–operator Storage 0.057 0.0180.542 156 156 × 1.7
combination and two with an- Apple Harvest 0.0084 0.0042
0.542 156 156 × 1.1
other device–operator combina- Storage 0.015 0.0031
0.542 156 156 × 2.6
tion. This methodology was zEff. = efficiency of estimation of means of device–operator combinations in the within-fruit stratum.
determined after preliminary yAlt. expt. = number of fruit required to achieve the same precision in an alternative experimental design that
studies with apples and kiwifruit tests different device–operator combinations at the between-fruit stratum with four measurements per fruit.
using one device (Instron). There Thus, for 1991, kiwifruit at harvest, the relative number is (0.556 × 0.40) ÷ 0.024 = 9.3.
are three comparisons of pen- xDegrees of freedom.
etrometer measurements x1, x2,
x3, x4 (taken in order around the fruit equator). One (x1 + x3 – x2 – differences in variability between kiwifruit and apples and be-
x4) compares pairs of opposite measurements, while the other two tween harvest and storage. These are average variabilities over
(x1 + x2 – x3 – x4) and (x2 + x3 – x1 – x4) compare pairs of adjacent devices and over operators. Table 3 also includes information that
measurements. The variance of these comparison statistics mea- allows an assessment of the benefit of our design relative to a
sures the consistency of the pairs whose difference is taken, with design in which different device–operator combinations were
small variance indicating high consistency. The variances were always tested on different fruit. To keep the comparison fair, we
28.3, 42.2, and 37.5 for apples and 42.2, 85.6, and 70.5 for assume, for both designs, four measurements per fruit. The table
kiwifruit. While the differences were not significant (P = 0.05), shows how, depending on fruit-to-fruit consistency, an experi-
and outliers were a contributing influence, these results provided mental design that allows comparisons to be made at the within-
a prima facie case for working with pairs of opposite measure- fruit level of variation may reduce the number of fruit and
ments. The plots shown in Fig. 3 provide a useful visual check on measurements needed to achieve a given precision dramatically.
the consistency between different pairs of readings. The first plot, Using a better design often is preferable to increasing the number
comparing pairs of opposite measurements, shows less scatter than of experimental units. Where fruit were more consistent (compare
the remaining plots, which compare pairs of adjacent measure- apples with kiwifruit; Table 3) the gains are smaller.
ments. Differences between methods. Measurements of firmness are
Fruit variability. Within-fruit mean squares (variances) for used in two ways. First, the absolute firmness value can be used in
different devices on the same fruit in Table 3 gives an indication of a regulatory fashion to define the minimum standard of fruit that

J. AMER. SOC. HORT. SCI. 121(5):927–936. 1996. 931

Fig. 4. Fruit firmness according to measurements by Instron, EPT pressure tester, and Effegi penetrometer operated by hand or using a motorized press. Differences
between operators are indicated for the hand-held Effegi penetrometer and EPT pressure tester. Measurements were made during the 1991 fruit season using apples
(A) at harvest and (B) after storage and kiwifruit (C) at harvest and (D) after storage. EPT = EPT pressure tester, Hand = hand-held Effegi penetrometer, and M. Press
= motorized press.

932 J. AMER. SOC. HORT. SCI. 121(5):927–936. 1996.

Fig. 5. Fruit firmness according to measurements by Instron, twist test, and Effegi penetrometer operated by hand or using a drill press. Differences between operators
are indicated for the twist test (except with apples, see text) and Effegi penetrometer operated by hand or using a drill press. Measurements were made during the 1992
season using apples (A) at harvest and (B) after storage and kiwifruit (C) at harvest and (D) after storage. Hand = hand-held Effegi penetrometer, Drill = Effegi
penetrometer operated using a drill press, and Twist = twist test.

J. AMER. SOC. HORT. SCI. 121(5):927–936. 1996. 933

growers must achieve before their fruit will be accepted by export general, the device for which the variance is smallest is clearly
or retail organizations. Second, change in firmness is used to best. Outliers, which may be due to occasional soft or hard spots,
characterize the rate at which fruit soften. In our study, measure- should ordinarily be excluded in the comparison. The robust
ments of firmness varied according to the device used. With hard estimates effectively exclude outliers. Frequent outliers, however,
fruit (apples at harvest, Fig. 4a and 5a) all devices except the may indicate that the device is prone to giving aberrant readings.
motorized press provided similar estimates of firmness once Three of eight Instron variance estimates were ≈0.03, and much
differences between operators were excluded. The motorized larger than variances 0.0017 to 0.018 estimated for other devices
press (Fig. 4a) underestimated firmness. With soft fruit (stored (Table 4). Using robust estimates reduced all these estimates by a
apples, 60 to 69 N and kiwifruit at harvest and after storage) (Fig. factor of three or more. The Instron may have been prone to
4), the EPT pressure tester provided the highest estimate of occasional aberrant readings due to the way fruit were held by hand
firmness followed by the hand-held Effegi penetrometer, motor- (perhaps the fruit twisted during penetration of the probe). Using
ized press, and the Instron, respectively. Bongers (1992), who S-PLUS’s (Statistical Sciences, 1991) default boxplot criterion for
assessed firmness of apples (average firmness between 50 and 58 outliers, average numbers of outliers per fruit in differences
N, which is equivalent to soft apples in our study), using the EPT between the two reading were Instron 9%, motorized press 8%,
pressure tester, Instron, and Effegi penetrometer mounted in a drill hand-held Effegi penetrometer 3%, Effegi penetrometer operated
press found a similar ranking of devices. The results from both by drill press 6%, twist test 5%, and EPT pressure tester 6%. A χ2
studies indicate the need for standardization in any regulatory use test suggests that the outlier rate does differ between devices (P =
of firmness measurements. Differences in machinery and operator 0.05). For each of the columns in Table 4, the Instron robust
may considerably affect the measurement of absolute firmness. variance estimate was always largest or close to largest.
Ability to discriminate between hard and soft fruit. Examining Comparing fruit after harvest and after storage. A comprehen-
Fig. 4 and 5 may suggest that the Instron is better than other devices sive comparison would take into account the effect of operator-to-
at detecting differences in firmness. For example, the Instron gave operator variability and assess ability to discriminate when presented
a 14.8-N decline in apple firmness in 1991 (Fig. 4a; cf. Fig. 4b). with a wide range of firmness. One such test is the ability to
Measurements by EPT pressure tester and Effegi penetrometer discriminate between fruit at harvest and after storage. For this
operated by hand or motorized press indicated average firmness comparison, we divided differences in firmness estimates between
declined an average of 9.3, 11.2, and 8.1 N, respectively. However, harvest and storage by a SED that was based on fruit-to-fruit
while the firmness difference is larger for the Instron, this differ- variability (i.e., on the between-fruit mean square of Table 3). This
ence was often offset by the lower precision with which this gives a t statistic, which we used as a measure of how well the
difference is measured. Differences between measurements on device discriminates. For comparing t statistics, a difference of ≈2
opposite sides of one fruit provide one possible measure of the is significant at P= 0.05, and a difference of 2.8 is significant at P
inherent precision of equipment. We checked variance estimates = 0.01. Results from this analysis are presented in Table 5.
for the possible effect of outliers. Table 4 gives nonrobust and, The most precise measurements of change in firmness accord-
where these differed by >15%, robust variance estimates. In ing to calculated t statistics were obtained using the Instron and
hand-held penetrometer for
Table 4. Within-fruit variance estimates. Devices that give small variances, and hence are more precise, are to apples in 1991, the Instron for
be preferred. Where the robust estimate of variance is >15% less than the nonrobust estimate, this is given apples in 1992, the hand-held
also. Robust estimates omit any “outliers” from the calculation. penetrometer for kiwifruit in
1991, and the twist test for kiwi-
Harvest Storage fruit in 1992. Note that the hand-
Instrument 1991 1992 1991 1992 held penetrometer was among
Kiwifruit the more precise devices. The
Instron 0.013 (18 df) 0.030 (24 df) 0.034 (18 df) 0.0095 (24 df) benefit of the Instron may be that
(0.010) z
(0.0095) z it is independent of operator.
Motorized press 0.018 (18 df) --- 0.0053 --- Influence of operator differ-
(0.011)z ences. The importance of opera-
Hand-held 0.0067 (72 df) 0.010 (96 df) 0.0051 (81 df) 0.0067 (105 df) tor differences should be
(0.0049) z
(0.0038)z considered relative to the abso-
EPT pressure tester 0.0072 (72 df) --- 0.0064 --- lute changes in firmness that are
(0.0049)z likely to occur during storage,
Drill press --- 0.011 (96 df) --- 0.018 (0.011) z the relative firmness differences
Twist test --- 0.0054 (96 df) --- 0.0084 that are likely to occur between
treatments, or both. Thus, differ-
Apple
ences between operators of ≤2.16
Instron 0.0033 0.0035 0.0035 0.030 N associated with hand-held
(0.0073)z Effegi penetrometers (Fig. 4d)
Motorized press 0.0022 --- 0.0024 --- are of relatively minor impor-
z
(0.00090) tance when compared to the 85-
Hand-held 0.0024 0.0021 0.0017 0.0024 N decline in firmness that can
(0.0013)z occur during kiwifruit softening
EPT pressure tester 0.0025 --- 0.0018 --- (Beever and Hopkirk, 1990).
Drill press --- 0.0034 --- 0.0040 However, apples soften only
Twist test --- 0.0025 --- 0.0028 slightly during storage and dif-
z Robust estimate of variance. ferences between operators of

934 J. AMER. SOC. HORT. SCI. 121(5):927–936. 1996.

≤7.6 N are relatively large when compared to the 14.4-N decline Table 5. Precision with which devices discriminate between hard and soft
in firmness (from 74.9 to 60.5 N; Instron, Fig. 4 a and b). fruit. Values represent t statistics, which in this context, we use to
To eliminate operator differences, it is recommended that the measure how well the device discriminates between fruit at harvest and
same operator be used for all penetrometer measurements (Blanpied fruit after storage. The higher the t statistic the better the precision. See
text for full description.
et al., 1978). The data presented in our study (Fig. 4 and 5) and in
earlier studies (Blanpied et al., 1978; Voisey, 1977) confirm that Harvest
using different operators can significantly influence firmness
Instrument 1991 1992
values. Bongers (1992), however, did not detect any difference
Kiwifruit
between the four operators used in his study.
All hand-operated devices examined in this study were subject Instron 6.1z 19.2
to operator differences, with the magnitude of the difference Motorized press 10.7 ---
varying between devices and between soft and hard fruit (Figs. 4 Hand-held 9.6–16.6y 18.9–28.4
and 5). Operator differences associated with the EPT pressure EPT pressure tester 10.9–15.0 ---
tester were minimal on firm apples but more substantial on soft Drill press --- 19.1–26.1
apples and kiwifruit (Fig. 4). With hard apples, operator differ- Twist test --- 25.1–26.1
ences were greater when Effegi penetrometers were operated Apple
using a drill press than when operated by hand (Fig. 5a). This Instron 8.7 13.5
ranking reversed with soft apples and kiwifruit (Fig. 5 b–d). We Motorized press 6.2 ---
found that the drill press was more awkward and slower to use, and Hand-held 5.2–11.2 5.4–9.6
did not always reduce operator differences. Before proceeding EPT pressure tester 5.6–7.8 ---
with using a drill press, it is worth checking that this method is Drill press --- 3.6–6.5
providing some advantage over the hand-held penetrometer. Dif- Twist test --- –4.1–5.1
ferent results might be expected depending on the model of drill z Singlet-statistics are given where device is independent of operator.
press used and on the strength of the springs and smoothness of the yThe range of t statistics is given when multiple operators have used the
downward action. device.
The cause of operator differences associated with hand-held
penetrometers may arise from the speed, angle, depth, or accelera- consistency is an interesting issue, particularly if measurements of
tion used to push the probe into the fruit. In preliminary investiga- firmness by a single operator vary from day to day depending on
tions using the Instron, we found that the speed and angle at which his or her mood and health.
the probe enters the fruit appears to have little effect (data not Ease of use of devices. Hand-operated devices such as the Effegi
shown). This is in agreement with recommendations that the penetrometer give full manipulative control of position of pen-
Instron can operate at speeds between 50 and 250 mm·min–1 when etrometer and fruit. When the Effegi was mounted in a drill press
measuring flesh firmness (Smith, 1985). Examining force– and a motorized press, the operation was slowed by the require-
distance curves generated by the Instron (data not shown) indi- ment for accurate placement of the fruit. All Effegi penetrometer
cated that force reached a maximum value within the first few measurements involved reading the dial and entering the value
millimeters of penetration in this study and, subsequently, de- onto paper or directly into a personal computer. This increases the
clined or remained constant (type B and C curves; Bourne, 1982). possibility of errors occurring during reading and recording of
Thus, operator differences are unlikely to be caused by differences values. In our opinion, the EPT pressure tester was the easiest
in the depth that the probe is pushed into the fruit. device to use to collect firmness data. Fruit were easily positioned
Comparisons of measurements by the motorized press and under the probe, the movement of the hinged arm was simple, and
Instron suggest that operator differences may be associated with the data was automatically transferred to a computer or printed
the way the probe was accelerated into the fruit. The motorized onto paper.
press was subject to a systematic error. While the penetrometer When considering the ease of use, it is important to examine
was driven downward at a constant speed, acceleration into the how the measurement of firmness fits in with other tests of fruit
fruit was determined by the distance the spring was compressed quality. For example, evaluating apple maturity and quality at
before the probe entered the fruit. Thus, we speculate that accelera- harvest routinely involves measurements of internal ethylene
tion into the flesh was slow in hard fruit and rapid in soft fruit. concentration, skin background color, flesh firmness, soluble
Comparisons of the motorized press with the Instron, which has a solids concentration, and starch pattern index (e.g., Watkins et al.,
constant velocity, indicate that the motorized press underestimated 1992). In our laboratory, we often combine measurements of
firmness of hard fruit (Fig. 4a) and overestimated firmness of soft firmness and soluble solids by transferring juice expressed onto the
fruit (Fig. 4d). Voisey (1977) found that, while most operators tip of the penetrometer probe directly onto a refractometer (Watkins
apply a smooth increase in force when using hand-held penetrom- et al., 1992). This operation is easily performed when using the
eters, other operators changed the rate of force application or Effegi penetrometer operated by hand, Effegi penetrometer mounted
momentarily relaxed the force just as the tissue yielded. in a drill press, and the EPT pressure tester.
Ranking of operators according to firmness measurements gave
a similar order on each assessment date. For example, Amy always Conclusion
provided the highest firmness measurement with a hand-held
Effegi penetrometer (Fig. 4). Thus, it may be reasonable to assume This study shows the large gains that are available from an
that operator-to-operator differences are similar at all evaluation effective experimental design. It demonstrates that differences
times. Further studies are required to validate this finding. Our between operators can be a significant problem when hand-
experiments involved relatively small numbers of measurements operated devices are used to measure firmness. Our results rein-
for any operator. We have no information whether operators can force the recommendation that the same operator be used when
maintain consistency over a long series of measurements. Operator hand-operated penetrometers are used to measure flesh firmness

J. AMER. SOC. HORT. SCI. 121(5):927–936. 1996. 935

for comparisons of pre- and postharvest treatments or to follow Kendall, M.G. and A. Stuart. 1979. The advanced theory of statistics, 4th
fruit softening (Blanpied et al., 1978). When devices, such as drill ed. Griffin, London.
presses, are used to drive the penetrometer probe into fruit, it would Magness, J.R. and G.F. Taylor. 1925. An improved type of pressure tester
be sensible to check that they do indeed improve the precision of for the determination of fruit maturity. U.S. Dept. Agr. Dept. Circ. no.
350.
measurement and reduce differences between operators over the Maindonald, J.H. 1992. Statistical design, analysis, and presentation
more simple devices available. Using a materials testing machine issues. N.Z. J. Agri. Res. 35:121–141.
to drive the penetrometer probe into fruit seemed to provide a Miller, R.G., Jr. 1986. Beyond ANOVA, basics of applied statistics.
consistent method of measuring firmness. With apples the Instron Wiley, New York.
provided a precise measurement of softening. Although the Instron Payne, R.W., P.W. Lane, P.G.N. Digby, S.A. Harding, P.K. Leech, G.W.
provided less precise measurements of kiwifruit softening, it has Morgan, A.D. Todd, R. Thompson, G. Tunnicliffe Wilson, S.J. Welham,
the benefit of greater operator independence. and R.P. White. 1993. Genstat 5 Release 3 reference manual. Clarendon
Press, Oxford.
Literature Cited Pearce, S.C. 1983. The agricultural field experiment: A statistical exami-
nation of theory and practice. Wiley, New York.
Abbott, J.A., A.E. Watada, and D.R. Massie. 1976. Effegi, Magness- Saltveit, M.E., Jr. 1978. Selecting an experimental design for pressure
Taylor, and Instron fruit pressure testing devices for apples, peaches, and testing apples. J. Amer. Soc. Hort. Sci. 103:105–109.
nectarines. J. Amer. Soc. Hort. Sci. 101:698–700. Smith, S.M. 1985. Measurement of the quality of apples: Recommenda-
Beever, D.J. and G. Hopkirk. 1990. Fruit development and fruit physiol- tions of an EEC working group. Commission European Comm.
ogy, p. 97–126. In: I.J. Warrington and G.C. Weston (eds.). Kiwifruit: Statistical Sciences. 1991. S-PLUS reference manual, version 3.0. Statis-
Science and management. N.Z. Soc. Hort. Sci., Auckland. tical Sciences, Seattle.
Blanpied, G.D., W.J. Bramlage, D.H. Dewey, R.L. LaBelle, L.M. Massey, Studman, C.J. and Yuwana. 1992. Twist test for measuring fruit firmness.
Jr., G.E. Mattus, W.C. Stiles, and A.E. Watada. 1978. A standardized J. Texture Studies 23:215–227.
method for collecting apple pressure test data. New York’s Food and Voisey, P.W. 1977. Examination of operational aspects of fruit pressure
Life Sciences Bul. 74:1–8. tests. Can. Inst. Food Sci. Technol. 10:284–294
Bongers, A.J. 1992. Comparison of three penetrometers used to evaluate Watkins, C. and J. Harman. 1981. Use of penetrometer to measure flesh
apple firmness. Washington State Tree Fruit Postharvest J. 3:7–9. firmness of fruit. Orchardist N.Z. 54:14–16.
Bourne, M.C. 1982. Food texture and viscosity: Concept and measure- Watkins, C.B., I.J. Horner, and I.B. Ferguson. 1992. The effects of collar
ment. Academic, San Diego. rot (Phytophthora cactorum) on “Cox’s Orange Pippin” fruit quality. J.
Chambers, M.C. and T.J. Hastie. 1991. Statistical models in S. Wadsworth Hort. Sci. 67:495–500.
and Brooks/Cole, Pacific Grove. Worthington, J.T. and J.N. Yeatman. 1968. A statistical evaluation of
Gacula, M.C. and J. Singh. 1984. Statistical methods in food and con- objective measurement of apple firmness. Proc. Amer. Soc. Hort. Sci.
sumer research. Academic, Orlando, Fla. 92:739–747.

936 J. AMER. SOC. HORT. SCI. 121(5):927–936. 1996.