Previous Page Table of Contents Next Page

Loss assessment using intervention load tracking


G.R. Akande[6], D. Jeffries[7] and A.R. Ward[7]


Load tracking was developed to provide statistically valid estimates of post-harvest fish losses. This paper shows how load tracking can be extended to compare loss-reducing interventions with the incumbent method. The statistical design issues are illustrated through the description of an intervention load tracking performed in Nigeria. An initial load tracking showed that there was a significant loss from cartons of fish during loading, transport and unloading. An appropriate intervention was the lining of the cartons with plastic bags and an intervention load tracking was designed to compare the loss from lined and unlined cartons. The analysis of the intervention load tracking experiments, must reflect the multi-level structure of the design. The major issues of the analysis are demonstrated using the data from the intervention load tracking performed in Nigeria. Both physical and quality loss were analysed and it was shown that the intervention reduced both types of loss, but only quality loss (i.e. breakage) was significantly (at the 5% level) reduced by the intervention.


Post harvest fish losses are associated with poor capture and post-capture activities. Reducing loss provides better benefits for the producer as well as the consumer. Methods to better understand losses provide those interested in loss reduction with a basis for informed decision-making at the micro and macro levels. Collaborative research by WADAF's West Africa Regional Programme for the Post Harvest Utilization of Artisanal Fish Catches and NRI's Post Harvest Fisheries Research Programme produced three field-based loss assessment methods which make it possible for investigations to be made and appropriate interventions devised for a variety of circumstances (Ward and Jeffries, 2000).

One of the methods, Load tracking was further developed to provide statistically valid estimates of post-harvest fish losses. It is a method which tracks losses of fish between stages or activities in a distribution chain (Ward and Jeffries, 2000). Random sampling is used to make the loss estimates representative of a wider population and adequate replication is used to provide accuracy measures for the loss estimates.

Load tracking experiments have been successfully planned and performed in a number of countries. For those processing chains that showed an important loss a natural follow-up is to find methods for reducing the losses. (The concept of important losses here combines statistical definition of significance with the economic implications).

This may result in various interventions being proposed.

This paper considers the design of experiments to compare interventions with normal practice. The general structure and design issues of intervention experiments are considered. A specific example describes the design of an intervention load tracking performed in Nigeria, where an intervention was used to reduce loss of smoked fish during transport by lorry between markets. It should be noted that this intervention was used to identify more specifically why loss was occurring and is not necessarily the recommended intervention under the circumstances.

Although the paired structure of conventional load tracking designs is present there is a further level for intervention load tracking experiments, comparing the interventions and the original method. This is a multi-level (Clark and Kempson, 1997) and the analysis must reflect this structure. A complete analysis of the data from the Nigerian intervention load tracking complements a general discussion of the analysis procedures.

Loss has various definitions (Ward and Jeffries, 2000), but load tracking is most appropriate for quantifying loss that can be measured on a continuous scale. The most common measure for loss is weight, which can obviously be used to quantify actual weight loss known as physical loss as well as the weight of fish that is downgraded, known as quality loss (Ward and Jeffries, 2000). It can also be employed to quantify quality loss by considering the weight of broken pieces of smoked fish. The data from the Nigerian intervention load tracking considers actual weight loss and the increase in weight of downgraded fish (broken pieces), during loading, transport and unloading of traditionally produced smoked fish. Earlier work (Akande and King, 1998) on load tracking of smoked West African Sardines (Sardinella maderensis) using the same considerations for quantifying quality loss showed that losses occurred during loading, transportation and unloading.


In common with any experimental design key areas for consideration are, objectives, what response to measure, and definition of the experimental unit, random sampling and replication.


The objectives do not need to be quoted as hypotheses and simple statements are perfectly acceptable. Typical objectives of intervention load tracking experiments are:

What response to measure?

Previous work on post-harvest fish loss assessment has identified four key types of loss: physical, quality, market force and nutritional (Ward and Jeffries, 2000). Load tracking can be used to quantify any of these losses, but requires a quantitative and non-subjective measure of loss such as weight and has been used to quantify physical and quality loss.

Ideally, the researcher should stay with the sampled units through out the experiment, to record a full post-harvest history including if possible data on time and temperature. This will strengthen the inference from the experiment enabling assumptions to be made concerning causes of loss. For practical reasons it might be people working in the distribution chain who record the post-harvest history.

One very important consideration when deciding on a response, is that the physical process used to measure loss must not its self incur losses.

Random sampling

The purpose of a designed experiment is to provide data suitable for statistical analysis from which information about a larger population can be inferred. This wider inference relies on the fact that the data has been obtained from random sampling. It is often necessary for practical reasons to use a two stage sampling process with a systematic selection followed by random sampling. Common sense should prevail in the choice of a sampling scheme, as there is no definitive method.

Experimental unit

In common with any experiment, the definition of the experimental unit is extremely important. It would be unwise to use one fish as an experimental unit as any differences detected are likely to be influenced by differences between the individual fish. For many load-tracking experiments, a container of fish makes a natural choice for the experimental unit. However, there are situations where this is not feasible or practical and the experimental unit will consist of a randomly sampled group of fish. It is difficult to give prescriptive estimates for the sample size, as it will be influenced by the size of the fish, but whatever the size each sample should contain two or more fish.


The purpose of LT is to obtain an accurate estimate of losses. The precision of an experiment is governed by the amount of replication of the experimental unit. Although there is no prescriptive rule for determining the correct level of replication a number of guidelines can be given. The major issues are:

Sensible estimates for the amount of replication are design dependent.


All load tracking experiments are based on the concept of identifying an experimental unit, assessing loss at the beginning of the tracking and then measuring the loss on the same experimental unit when the tracking terminates. The continuity of the same sample is what defines the concept of load tracking.

The before and after measurements are both taken within the same experimental unit and at its simplest level the design has before/after as a treatment factor and the experimental unit as a blocking factor. This is more commonly known as a paired t-test. However an intervention load tracking is more complicated, as there is now a ‘treatment' level where the interventions and control (normal practice) are compared.

The sampling units for the intervention load tracking designs in this paper are taken to be whole containers of fish. There may be cases when this is not convenient and sub-sampling of containers would be necessary. This raises further design issues and will not be considered in this paper.


The design of the experiment must consider, that the before and after measurements are taken within treatment for each container, which gives a design with two levels of variation. This must be acknowledged in the design and analysis and can be thought of as a split-plot type approach.

To ensure that all interventions and control are compared fairly, it is advisable to use blocking for intervention designs. Each block should consist of a full replicate of containers, i.e. one container for each intervention and one for the control. Physically this means that these containers should be subject to similar conditions. So, for example if the intervention load tracking was looking at storage interventions, the containers should be stored close together so that they all experience similar conditions.

Figure 1. Multi-level design

Main level

Split level



Intervention 1

Before Intervention 1

After Intervention 1

Intervention 2

Before Intervention 2

After Intervention 2










Intervention n

Before Intervention n

After Intervention n

Generally blocking should always be used. If blocking was not used, but turned out to be required, the inference from the load tracking is likely to be false. The control/interventions levels can be thought of as the main plot level and before/after as the split-plot level. This two level approach is illustrated in Figure 1 for a single block.

Figure 2. Interaction effects

Figure 2a. Little evidence of interaction

Figure 2b. Evidence of interaction

The objective of the experiment is to compare the losses for the control and the interventions. This is measured by the interaction effect in the split-level of control and interventions and before/after. The comparison has to be performed in the split-level, as there is no information about the before/after comparisons in the main plot.

Figure 2 considers two examples of interaction effect for the weight of broken pieces, where there is assumed to be a control (standard practice) and an intervention.

In Figure 2a, the increase in the weight of broken pieces is similar for both the control and the intervention, hence almost parallel lines. However in Figure 2b, there is greater evidence of an interaction with time, as the increase for the control containers is larger.


Figure 3. Skeleton split-plot analysis

Source of variation

Degrees of



Control and intervention


Main plot residual


Before vs. After


Interaction of (control & intervention) with (before & after)


Split-plot residual


Total 2rt - 1

Since the difference in losses is quantified by the interaction effect in the split-plot level, the replication should be chosen to ensure adequate degrees of freedom for the residual in this level. A sensible level of replication should give between 10 and 20 degrees of freedom for the residual.

Considering the structure of the skeleton analysis of variance table can choose the amount of replication. For an intervention load tracking with t treatments (1 control, t-1 interventions), each replicated r times. Figure 3 shows the skeleton analysis of variance with the degrees of freedom for each effect.

The number of interventions to be compared, will be known for an intervention load tracking and the number of replicates, r, can then be calculated by ensuring that the expression t(r-1) is at least double figures. For example, for one control and one intervention there should be at least six replicates. For a fair comparison, each replicate would be physically arranged so that each intervention and the control were subject to similar conditions. Here, the replicate is being used as a blocking factor, which helps to increase the precision of the load tracking


An important sampling issue in intervention load tracking is the filling of the containers. If all the intervention containers are filled first and then all the control containers, the results of the analysis are likely to be biased. If the containers are filled in random order an unfortunate ordering could again result leading to bias. The most appropriate procedure in this case is to consider a systematic alternate filling of control and intervention container.


A conventional load tracking experiment (Ward and Jeffries, 2000) was performed in Nigeria in March 2000, to investigate the losses incurred by loading, unloading and long-distance transport by lorry from Maiduguri to Lagos.

The results of this load tracking suggested that lining the fish container with a plastic bag might reduce losses and a further intervention load tracking experiment was undertaken to test this theory. The reduction may result from the prevention of:

Maiduguri has a large wholesale fish market where the containers of fish bound for Lagos are called ‘Lagos' cartons. A sample size of 16 Lagos cartons (eight each for the control and the treatment) of smoked fish was chosen to give a compromise between precision and cost. Sixteen cartons gives fourteen degrees of freedom for the estimate of the split-plot variance.

The Nigerian Institute for Oceanography and Marine Research (NIOMR) undertook this experiment in May 2000, with assistance from the Federal Department of Fisheries (FDF) and Borno State Fish Sellers Association, Maiduguri.


The purpose of the intervention load tracking experiment was to compare the losses between lined and unlined cartons. Two types of loss were considered, actual weight loss (physical loss) from the fish stored in the cartons and quality loss, quantified by the increase in weight of broken pieces. The stages of the distribution chain tracked were the packing, loading, transport and unloading of fish, with no separation of the constituent parts.


Three grades of whole smoked catfish fish (Clarias gariepinus) are sold in Maiduguri market, Grade 1 being the best quality and most expensive, grade 2 of medium price and quality with grade 3 having the poorest quality and lowest price. Grade 2 medium sized fish was chosen for this exercise, because it was the most commonly available product.

There were many traders from whom fish could have been bought, but two were randomly chosen. Enough fish to fill 16 Lagos cartons were purchased and then heaped on a tarpaulin sheet laid on the ground. Two market workers randomly selected by the researchers filled the cartons in a traditional manner. The plastic linings of the interventions were sealed. After filling, the cartons were tied with rope according to local practice. The cartons were numbered 1 to 16 with a marker pen, and the numbers of the intervention and control cartons was recorded.

Normally, the next stage would be to either store the packed cartons or load them onto a lorry for onward transport to Lagos. However, for the load tracking exercise the cartons were carefully unpacked and the weight of the carton, the fish and the proportion of broken fish were measured using a dial weighing scale. Market workers, as normally practised sorted the broken fish from the whole fish. The fish, including the broken pieces were re-packed and the cartons stored ready for transport. Great care was taken not to introduce losses during the unpacking, weighing and re-packing stage. Undertaking a load tracking of this stage to assess any losses introduced could eliminate any doubt about this.

To ensure that the control and intervention cartons were subjected to similar conditions during the journey to Lagos, the cartons were arranged as eight pairs (i.e. blocks), with each pair consisting of one control and one intervention carton. The pairs were systematically assigned to separate levels in the lorry, but within each level a pair was randomly located. This reduces the chance of bias, which would be caused by locating all the experimental cartons in a small number of levels. The cartons were loaded in the usual manner with minimum input from the researchers.

Figure 4. Timetable of key activities for intervention experiment






Purchase of smoked fish



Mixing and packing



Unpacking of cartons and grading into whole and broken to assess loss Cartons repacked and roped.



Cartons kept for loading


2hr 45min

Loading onto lorry




Lorry leaves for Lagos



Lorry arrives Lagos




Unloading of cartons



Unpacking of cartons and repeat loss assessment as carried out in Maiduguri



The lorry was packed with 900 cartons, mainly of the "Lagos" type, which constituted a full load. The lorry was loaded on a Thursday and left Maiduguri the next day. It arrived in Lagos on Saturday during the night and was unloaded on Sunday morning. The unloading and weighing procedure was exactly the same as in Maiduguri, with each carton being carefully emptied onto a tarpaulin sheet. Market workers then sorted the fish into whole and broken categories. It is of course impossible to ensure identical grading of fish in Maiduguri and Lagos, but it was clear that the methods were consistent. As in Maiduguri the relevant weights were recorded (the weights of the bags were deemed negligible). After this procedure the experimental cartons were repacked and sold to recover part of the expenses, hence reducing the cost of the experiment.

The timetable of the key events in this intervention load tracking is given in Figure 4.


In common with any analysis an important tool is the appropriate use of graphical methods and summary statistics. Formal analysis for intervention load tracking is centred on analysis of variance and the presentation of confidence intervals for appropriate comparisons. However as demonstrated in this section there are some issues that must be considered and a prescriptive approach for intervention load tracking is not justified.

Analysis of weight of fish in carton

Figure 5 gives the before and after weights of fish in the control and intervention cartons. There is an indication that weight loss is less for the intervention cartons, but this should be tested formally.

The split-plot analysis of variance (note the constant variance and normality assumptions were satisfactory) is shown in Figure 6, where the highlighted row shows no significant interaction at the 5% level.

Figure 5: Comparison of before and after cartons

Before and after weights for control

Before and after weights for intervention

Figure 6: Split-plot analysis of variance for weight loss from carton.

Source of variation





F pr.

block stratum





main plot

control vs. intervention











split plot

before vs. after



















(d.f. = degrees of freedom, s.s. = sum of squares, m.s. = mean of squares)

Note the main plot effects are averaged over time and have no relevant interpretation. The before vs. after effect is averaged treatment and again has little relevance to the analysis.

The interaction is not significant, indicating that the losses for the control and intervention cartons are similar. The losses can be summarised by calculating confidence intervals (using the variance estimate from the analysis of variance) for the loss from the control and intervention cartoons, as shown in Figure 7.

Figure 7. 95% confidence intervals for weight loss.


Estimate of mean
weight loss per

95% CI for mean
weight loss per

As %
weight loss


0.29 kg (2.9%)

0.18 to 0.41 kg

1.8% to 4.2%


0.17 kg (1.8%)

0.06 to 0.28 kg

0.6% to 2.9%

The intervention carton has less weight loss, but the confidence intervals overlap, confirming the non-significant result obtained for the interaction. The interaction is not even significant at the 10% level (p-value = 0.12 from Figure 6) and is unlikely to have economic importance.

Random sampling should ensure the weights of fish in the control and intervention cartons are roughly similar. In this case the average weight of fish in the intervention and control cartons was 9.7 kg and 9.5 kg respectively. Figure 8 shows that there is a trend for increasing percentage weight loss for increasing weight of carton.

Figure 8: % weight loss against initial weight of fish in carton

As the weights of control cartons are on average higher than those for the intervention the loss estimates for the control could be overestimates and those for treatment underestimates. Consequently, the comparison might not be ‘fair'. This can be allowed for by analysing the weight losses (weight before - weight after) and using the initial weight loss as a covariate. Figure 9 shows the unadjusted mean weight losses and the mean weight loss adjusted for the covariate effect of differing initial weight between the control and the treatment cartons.

Figure 9: Comparison of adjusted and unadjusted weight losses

Mean weight loss



Unadjusted (raw)

0.29 kg

0.17 kg


0.27 kg

0.19 kg

As expected, the control mean weight loss has been adjusted down and the intervention mean weight loss has been increased. However the changes are very small and in this case allowing for the covariate effect there is no significant difference at the 5% level between the control and the intervention losses. In this case the inference is identical to the split-plot analysis and given effective random sampling this will generally be the case.

Analysis of changes in broken fish weight

The same split-plot analysis (again the constant variance and normality assumptions were satisfactory) was applied to the weight of broken fish, which showed a significant (at the 5% level) reduction in breakage for the intervention cartons. However in this case there was a strong covariate effect, with the cartons with the highest initial weight showing the least increase in breakage. The adjusted mean differences for control and treatment are shown in Figure 10.

Figure 10: Summary of adjusted means for the increase in broken fish


Mean weight

95% CI for mean weight

% weight


0.63kg (6.4%)

0.51 to 0.75 kg

5.2% to 7.6%


0.09kg (0.9%)

Not signif. at 5%


Control (before - after) vs. Intervention
(before - after)

95% confidence interval


0.29kg to 0.82kg


2.5% to 8.6%

The intervention has significant advantages over the control in reducing the amount of breakage. However the important difference should be qualified by considering the economic implications.

The load tracking showed that between 2.5% and 8.6% more fish was downgraded from whole to broken using the normal method.

Whole grade II fish are sold in Mile 12 market Lagos for X currency units per Kg and broken fish are sold for Y currency units per kg. So from this load tracking the smoked fish from the control cartons have a price loss of between 2.5%*(Y-X) to 8.6%*(Y-X) per Kg more than the smoked fish from the intervention cartons.

For an average carton weight of 10 kg, this gives a minimum price loss of 10*2.5%*(Y-X) per carton. The only cost of intervention is the price of the lining, which is a Z currency unit per carton. If 10*2.5%*(Y-X) is greater than Z, then the intervention is saving the trader at least 10*2.5%*(Y-X)-Z per carton.

Note for comparison with other load tracking experiments it is convenient to express all losses as a percentage of the initial carton weight. As all the carton weights were similar, analysis in terms of percentage was justified. However if this was not the case then alternative numerical permutation techniques (Manly, 1997) might be considered for the analysis.


The paper has shown how conventional load tracking can be extended to compare normal practice with appropriate interventions. A specific example has been considered in detail, demonstrating how the effect of an intervention can be quantified for two different loss responses (physical and quality).

Further work could examine the feasibility of using load tracking to compare multiple parts of a distribution chain in one experiment. This might allow the separation of such effects as storage, loading, transport and unloading. Instead of having a simple before/after factor this would involve a multi-level factor, one level for each stage. One problem with this type of design is that the measurement of loss at four consecutive stages might have a large cumulative effect.

This paper has concentrated on researcher led experiments with precise quantitative measurement, but for local processors and fisherfolk this approach is clearly not feasible. In this situation it would be appropriate to develop a less formal empirical intervention load tracking technique. This could then be used at a local level to identify important losses, which could then be discussed with extension staff.


Akande, G.R., Jeffries, D.J. and Ward, A.R. (2000). Fish Loss Assessment Research. 44 pp. [DFID Project 7008 "Post Harvest Fisheries Research Programme"]

Akande, G.R. and King, A.M. (1998). Application of Load Tracking in the distribution chain of smoked sardines (Sardinella maderensis). A case study of Magbon-Alade Fishing Community, Lagos State, Nigeria. 20 pp. [DFID Project 7008 "Post Harvest Fisheries Research Programme"]

Clarke G. M. and Kempson R.E (1997). Introduction to the design and analysis of experiments. Arnold, 1997

Manly B.F.J. (1997). Randomisation, bootstrap and monte-carlo methods in biology. Chapman & Hall, 1997.

Ward, A. R. and Jeffries, D.J. (2000). A manual for assessing postharvest fisheries losses. Natural Resources Institute, Chatham Maritime, Kent mE4 4TB, United Kingdom.

Development and sensory acceptability of crackers made from the bigeye grunt (Brachydeuterus auritus)


Modupe Abimbola King
Nigerian Institute for Oceanography and Marine Research (NIOMR)
Victoria Island, Lagos, Nigeria.


The bigeye grunt (Brachydeuterus auritus), present in a large biomass in the Gulf of Guinea is generally considered an under- utilised fish species. In an attempt to add value, it was used to complement cassava starch (Manihot esculenta) to produce fish crackers. Three levels of percentage fish content (40%, 50% and 60%) and three levels of percentage starch content (60%, 50% and 40%) were used in the formulations. Proximate analyses and sensory evaluations were carried out. The results showed that protein, fat and ash content increased with increased proportion of fish. The sensory evaluation tests showed that the most acceptable formulations for the crackers were obtained using 50% fish/50% starch followed by 40% fish/60% starch combinations. The linear expansion of the fried crackers increased with increased proportion of fish. Fish crackers production apart from its appeal in increasing protein intake, has the potential of a small regional snack factory in a developing economy.


The bigeye grunt (Brachydeuterus auritus), present in a large biomass in the Gulf of Guinea (Williams, 1968; Teutscher, 1979), is generally considered an under- utilised fish species (Talabi et al., 1983). Its domestic consumption is limited by a size problem, with a reported maximum length of about 25 cm, even at maturity, and absence of a defined method of utilization, it thus attracts a low commercial value.

The bigeye grunt, therefore meets many of the criteria desirable for value addition; it is abundant and of low cost. One of the ways to add value is by using it to complement low protein staples such as cassava (Manihot esculenta) to make fish crackers, since cassava starch has been known to produce crackers with excellent expansion properties (Yu, 1991). Cassava, with protein content less than 3%, do not provide adequate protein for human requirements even when ingestion exceeds caloric requirements (Badrie and Mellowes, 1992). In contrast, the amino acid content of fish and its essential amino acid balance are appropriate for human consumption. Therefore, in a formulated mixture an improved balance of amino acid may be obtained.

Crackers (keropok) are popular snack foods in Malaysia and other Asian countries (Yu, 1997). They are produced by gelatinisation of starchy dough that is shaped into different forms before drying. Apart from the two essential components, which are starch and water, other ingredients such as fish or other types of seafood are usually added to produce different types of crackers (Vasanti Nair et al., 1996). Cassava starch/big eye crackers properly formulated with high nutritional and sensory quality, has the potential of being the source of essential nutrients, especially for children and teenagers. In addition, a snack with high nutritional and sensory quality, properly formulated and adequately packaged, may be an appealing product for small regional snack factories in developing countries (Tettweiler, 1991).

There is, however, a need to test the suitability of big eye and cassava starch for the production of fish crackers in terms of taste and other sensory attributes. This is because taste will continue to be a driving force in the acceptance of a new food product in the competitive market place in the food processing industry (Osnabrugge, 1989).

The objective of this study was to test the acceptability of fish crackers prepared from the bigeye grunt and cassava starch using different formulations.


Preparation of fish crackers

A sample of bigeye grunt was obtained from a fishing company, Obelawo Facha, in Lagos. Gutting was done manually and the flesh was separated using a mechanical deboner (BADDER 694). The mince obtained was mixed with cassava starch at a fish to starch ratio of 40:60 (A1), 50:50 (A2) and 60:40 (A3). 1% sugar, 1.5% salt, monosodium glutamate and 20-30% water were added to the mixture. All the ingredients were thoroughly mixed mechanically until a smooth paste was obtained. The paste was moulded with the aid of metal moulds and then steamed for 90 minutes. The moulded pastes were cooled in cold water, to minimize shrinkage, and chilled overnight in a refrigerator (Profi-line LIEBHERR) operating at 1°C to 5°C. The chilled pastes were sliced into a thickness of about 2-3 mm and dried in the oven (GALLENKAMP Plus II oven) at 50°C for 10 to 12 hours until moisture content of 10% (±2%) was obtained. The dried slices were deep fried in vegetable oil. Frying made the slices expand and increase in size to obtain a low - density porous product known as fish crackers.

Measurement of the linear expansion of the crackers

Linear expansion was determined by measuring five lines drawn across each chip before and after frying (Yu et al.,1981). The lengths of individual lines were measured with the aid of a thread and a measuring tape. The measurement was replicated ten times. The linear expansion of the crackers is a measure of textural quality and was calculated from the equation:

LE = 100 (Lf -Lo) Lo-1

where Lo and Lf are the lengths (cm) of the lines before and after frying respectively (Vasanti Nair et al., 1996).

Sensory evaluation and proximate analyses

A trained panel consisting of 10 members was used to evaluate the colour, texture (assessed as crispness), flavour and overall acceptability of the fried samples. The panellists evaluated the samples independently and the tests were duplicated. The treatments were compared by ranking. The ranks were converted to scores, which were then subjected to analysis of variance. Treatment means were further subjected to Tukey's test to evaluate the difference between the samples (Larmond, 1977). This initial study was followed by large-scale market testing during exhibitions in various locations within and outside the Country.

Crude protein (N × 6.25) was determined by the micro-Kjeldahl method (Pearson, 1976), total lipid was estimated by petroleum ether extraction. Total ash was determined by ashing for 12 hours at 550°C, moisture content was determined by weight difference of samples dried in a vacuum oven overnight at 60°C.

The linear expansion, sensory attributes and nutrient content of the three different combinations were compared.


Table 1. Chemical composition (%) of dried crackers from bigeye grunt and cassava starch. Means of two replicates.






A1 (40% fish/60% starch)





A2 (50% fish/50%starch)





A3 (60% fish/40% starch)





Tables 1 and 2 show the results of the proximate analyses of the different formulations. The protein content of dried and fried crackers expectedly increased with increased proportion of big eye fish. The crude protein content of the dried samples was comparable to that obtained from crackers prepared by using Clupea leiogaster and tapioca/sago flours (Yu, 1977). Frying lowered the protein contents of the crackers in relation to the dried ones as a consequence of oil absorption during frying (Table 2).

Table 2. Chemical composition (%) and linear expansion (%) of fried crackers prepared from bigeye grunt and cassava starch. Means of two replicates.







A1 (40% fish/60% starch)






A2 (50% fish/50%starch)






A3 (60% fish/40% starch)






Crispness, the most important sensory attribute of crackers, is directly related to linear expansion. A linear expansion greater than 77% is required for an acceptable level of crispness (Siaw et al., 1985). The linear expansion ratings of the fried crackers indicated a rise in linear expansion with increasing fish content (Table2). Previous work on expansion using soybean/cassava (Badrie and Mellowes, 1992), wheat flour/wheat starch (Faubion and Hoseney, 1982) tapioca/rice starch (Yu, 1993) reported that increasing the protein content of the blends resulted in a decrease in the linear expansion of the extrudates. However in a study of soya, wheat, milk and egg proteins, it was found that milk protein tended to increase expansion volume while the other proteins decreased expansion (Chinnaswamy and Hanna, 1990). This was associated with the viscoelastic nature and the cross-linking ability of different proteins. As the degree of cross-linking increases, the amount of expansion during frying decreases (Chinnaswamy and Hanna, 1990). The increase in linear expansion with increased proportion of fish might be associated with myofibrillar proteins (particularly myosin) present in minced fish, which have the ability to form gel.

Table 3. Mean values for sensory attributes of fish crackers prepared from big eye and cassava starch.


* Sensory attributes





A1 (40%fish/60% starch)





A2 (50%fish/50% starch)





A3 (60% fish/40%starch)







*NS = Not significant
Means followed by different letters are significantly different at P£0.05 by Tukey's test.

Taste panellists found no significant difference in the flavour and crispness of the different formulations (Table 3). However, they rated the colour and overall acceptability of A1 (40% fish/60% starch) and A2 (50% fish/50% starch) significantly (P£0.05) better than that of A3 (60% fish/40% starch). They commented that the preferred samples had brighter colours. Favourable comments were obtained at exhibitions conducted within and outside the country.


Acceptable fish crackers have been produced using big eye and cassava starch. Laboratory sensory scores and preliminary large-scale sensory evaluation trials show favourable and encouraging response.

Low-value fish species would remain important in a developing country like Nigeria. Processing these species into value-added products using simple technology would not only increase their economic value, but also encourage the exploitation of these under- utilized resources thus contributing to poverty alleviation and income generation of the fisher folks. Fish crackers production, apart from its appeal in increasing protein intake, has the potential of a small regional factory in a developing economy. Further studies are now in progress to determine the exact nature of factors found in fish protein that appears to cause increased expansion in fish crackers with increased protein content.


Williams, F. (1968). Report on the Guinean Trawling Survey. Organization of African Unity Scientific, Technical and Research Commission. Vol 1: 378-382.

Teutscher, F. (1979). A review of potential catches and utilization of small pelagic fish. FAO TF/INT 298 (DEN) Rome.

Talabi S.O., Fetuga B.L. and Ologhobo, A. (1983). Utilization of ‘big- eye' fish (Brachydeuterus auritus) for fishmeal and fish protein concentrate production. A preliminary biochemical and nutritional evaluation. By-products, 335-338.

Yu, S.Y. (1991). Acceptability of fish crackers (keropok) made from different types of flour. ASEAN Food J., 6: 114-116.

Badrie, N. and Mellowes, W.A. (1992). Soybean flour/oil and wheat bran effects on characteristics of cassava (Manihot esculenta Crantz) flour extrudate. J. Food Sci., 57: 108-111.

Yu, S.Y. (1977). Fish crackers - poised for new image. INFOFISH Int. July/Aug: 46-52.

Vasanti Nair C.K., Seow, C.C. and Sulebele, G.A. (1996). Effects of frying parameters on physical changes of tapioca chips during deep-fat frying. Int. Sci. and Technol., 16: 51-58.

Tettweiler, P. (1991). Snack foods worldwide. Food Technol., 45:58.

Osnabrugge, W. (1989). How to flavour baked goods and snacks effectively. Food Technol., 43: 72-82.

Yu, S.Y., Mitchell, J.R. and Abdullah, A. (1981). Production and acceptability testing of fish crackers (keropok) prepared by the extrusion method. J. Food Technol.,16: 51-58.

Larmond, E. (1977). Laboratory methods for sensory evaluation of food. Publication 1637; Canada department of Agriculture.Ottawa, Canada.

Pearson, D. (1976). The chemical analysis of Foods. 7th ed. London and New York: Churchill- Livingstone.

Siaw, C.L., Idrus A.Z. and Yu, S.Y. (1985). Intermediate technology for fish crackers (keropok) production. J. Food Technol., 20: 17-21.

Faubion, J.M. and Hoseney, R.C. (1982). High temperature short- time extrusion cooking of wheat starch and flour. Effect of moisture and flour type on extrudate properties. Cereal chem., 59: 529-533.

Yu, S.Y., 1993. Effect of rice starch on linear expansion of fish crackers (keropok). Trop. Sci. 33:319- 321.

Chinnaswamy, R. and Hanna, M.A. (1990). Extruded snack food: relationship between ingredient chemical qualities and expansion properties. Int. Food Ingredients, 3:3-9.

[6] Nigerian Institute for Oceanography and Marine Research, Wilmot Point Road, Bar-Beach, P.M.B.12729, Victoria Island, Lagos, Nigeria
[7] Natural Resources Institute, Medway University Campus, Central Avenue, Chatham Maritime, Kent ME4 4TB, United Kingdom.

Previous Page Top of Page Next Page