PREPARATION OF A DRY PRODUCT FROM CONDENSED MENHADEN SOLUBLES Statistical Supplement WOODS HOI t OCEANOGRAPHIC INSTITUTION F£b 1 S ii)0/ WOODS HOLE, MASS. SPECIAL SCIENTIFIC REPORT- FISHERIES No. 1 94 UNITED STATES DEPARTMENT OF THE INTERIOR FISH AND WILDLIFE SERVICE United States Department of the Interior, Fred A, Seaton, Secretary U. S. Fish and Wildlife Service PREPARATION OF A DRY PRODUCT FROM CONDENSED MENHADEN SOLUBLES: STATISTICAL ANALYSIS OF THE DATA (Supplement to the Fish and Wildlife Service Research Report hS) by C. F, Lee and R. J. Monroe Special Scientific Report - Fisheries No© 19U Washington, Do C, December 1956 CONTENTS Page Introduction ••• •••••(>••••• • 1 Object of the statistical analysis 2 Introduction to the statistical analysis •••• 2 Simple correlations •••.••••••e**»«*e*o« 3 Multiple correlations ••o«e»*»** 3 Discussion of tests for significance •••••• o 3 Analysis of variance ••••••• •••<>••*. 3 Discussion ooooo**«***********«o****e** U Simple correlations related to the Suitability Factor • • • e • h Minor simple correlations •••• •••••••• k Multiple correlations •e******»o****o«e«*»« 6 Factors determining specific gravity •o***«o«o*e« 6 Factors affecting total dry solids ••••e«o****o* 6 Factors affecting viscosity ••e*»*«*»e«o**o* 6 Factors affecting moisture absorption ••••••••••• 6 Stu(fy of source factors of solubles by analysis of variance • • 11 Area effects on composition of solubles ••••••••eo 11 Effect of process •• o •••• 11 Effect of month of collection 12 Summary e*«**«*o***o*o***oo*«»o(>*ooo* 17 Conclusion •••oooo***o*oo*o**«**o*o**»e 19 TABLES Table 1» — Simple correlation values •••••••e«««*.«o 5 Table 2. —Multiple correlation with specific gravity •••oo«o 7 Table 3o — Multiple correlation vrlth dry solids » <, » o 8 Table h» — ^Multiple correlation with viscosity o*»»«o««ee 9 Table $, — Multiple correlation with moisture absorption o • • • • 10 Table 6. — Analysis of effects of areas, processes, and dates of collection .oataaocvaaooo** 13 Table 7»-~Process means c«e«o»«*e**e*«**o«*«« Ik Table 8.— Means by month of sample collection eoo«*««*«* 15 Table ^.—IJumber of observations in the area-process subclasses • 16 ABSTRACT A statistical analysis has been made, as a supplement to D, S. Fish and Wildlife Service Research Report k^, Preparation of a Dry Product from Condensed Menhaden Solubles, Simple correlation studies show that k of 10 chemical and physical characteristics are important in preparing suitable dry condensed solubles. These are ammonia, corrected protein content, total ash and water insoluble matter, and desirable limits are suggested for each. Multiple correlation studies of other selected data show that the specific gravity of condensed solubles is largely determined by its total ash and fat content. Dry solids content and the refractive index or spe- cific gravity are not correlated either separately or in combination. The study indicates that the variability in composition of condensed solubles does not account for either the extreme variation in viscosity of the condensed solubles or for differences in moisture-absorptive char- acteristics of the related dry solubles. By means of an analysis of variance it was shown that the solubles produced by plants in the South Atlantic and Gulf of Mexico areas differed in composition from the product of plants in the Central Atlantic area. Otherwise, there were not sufficient data to permit determination of the effect of type of plant or of the month of production upon the properties of the condensed solubles. PREPARATION OF A DRY PRODUCT FROM CONDENSED MENHADEN SOLUBLES; STATISTICAL ANALYSIS OF THE DATA (SUPPLEMENT TO THE U. S. FISH AND VflLDLIFE SERVICE RESEARCH REPORT 16) by C. F. Leei/ and R, J, Monroe=/ INTRODUCTION In the course of an investigation of the practicability of preparing a dry product from condensed fish solubles (50 percent solids), a number of commercially produced samples of condensed menhaden solubles were col- lected. The 20 plants producing these samples were located along the Atlantic Coast from New Jersey to Florida and along the coast of the Gulf of Mexico from West Florida to Texas o Most of the samples were drawn di- rectly from storage tanks, although some were tsiken from the transfer pipes as the product was being pumped from the evaporators. The sanqjles were collected from various areas during October 1952 and from July to November 1953. Thirty-two samples of condensed solubles were used to prepare pure diy solubles on a laboratory drum dryer. This process, the description of the dry solubles produced, and the composition and physical properties of the condensed solubles used as starting material for the dry solubles are de- scribed in detail in the U, S, Fish and Wildlife Service Research Report U5, Preparation of a Dry Product from Condensed Menhaden Solubles, The present paper is a supplement to Research Report 1^5 and consists primarily of a statistical analysis and an interpretation of some of the data presented in table 1, page 9 of that report. Information on the source of saii?)les and on the 12 physical and chemical tests used to compare these samples is contained on pages 13-16 of the Research Report, On page 12 of the Report, the procedure used to derive a "Suitability Factor" for each sample of dry solubles is described. This value is derived from various chspracteristics observed during the drying operation and from certain prop- erties of the dry product and is intended to give an objective comparison of the various dry solubles. The Suitability Factor permits an estimation of the suitability of the corresponding sample of condensed solubles to make a satisfactory dry product. 1/ Chemical Engineer, Fishery Technological Laboratory, U. S, Fish and Wildlife Service, College Park, Maryland 2/ Acting Head, Deparfauent of Experimental Statistics, North Carolina State College, Raleigh, North Carolina There was a considerable range in each set of data for all char- acteristics of the condensed solubles that were studied and for the Suitability Factors, To illustrate the magnitude of this variability and the correlation, or more properly, the lack of a marked degree of correlation of these values to the Suitability Factor, scattergrams were drawn in which the values for each characteristic were plotted against the values of the corresponding Suitability Factor, These are reproduced in the Research Report as figures 3 to 12, inclusive. Object of the Statistical Analysis The object of this statistical analysis and interpretation was to determine whether significant relationships existed between the variable factors determined for the samples of condensed solubles and their suita- bility for preparing dry solubles, as represented by the Suitability Factor, A similar study of simple and multiple correlations was made for various other interrelated data. In some instances, these were for data iirtiere correlations would be expected, for example refractive index and specific gravity to total dry solids. In most cases inspection of the scattergrams or experience had lead to assumption of certain correlations and the sta- tistical analysis was used to verify these assujttptions and test their significance. Some discussion of these assumed correlations may be found on pages 18 and 2k of Research Report li5, A secondary object of the analysis of data was to explore the pos- sibility that certain of the variables studied might be selected as indicators of lots of condensed solubles that would be generally suited for making dry solubles. It was evident from the scattergrams that no single variable was highly correlated, but certain selected combinations might give increased accuracy of prediction. Introduction to the Statistical Analysis The large volume of data made it in^jractical to cariy out the desired statistical work at the College Park Laboratory, It was done through a cooperative agreement with the Depcirtment of Experimental Statistics, North Carolina State College, Raleigh, North Carolina, Dr, R, J, Monroe supplied the data presented in tables 1 to 9, and has also included the introductory paragraphs immediately following to facilitate interpretation of the tables. The senior author had appended a discussion of a less technical nature of the data in the tables to emphasize practical aspects and implications. Simple correlation, multiple correlation and analysis of variance techniques were used to provide evidence of association between physical and chemical characteristics and to measure the effects of certain ex- ternal factors on these characteristics. - 2 - Simple Correlations The simple correlation coefficients were determined from the sample n ■ 32o Each simple correlation coefficient was tested for significance using the standard tables at 30 degrees of freedom. The results are shown in table 1, One asterisk indicates that the sample value exceeds the table value at the 5-percent level of significancej 2 asterisks, at the 1-percent level J and 3 asterisks, at the 0,1-percent level. Multiple Correlations The multiple correlations vary in degrees of freedom, depending on the number of variables included. In general, degrees of freedom " n - p - 1, •vrfiere p is the number of independent variables included. The data are tabulated in tables 2 to 5 inclusive. Discussion of Tests for Significance The testing of either simple or multiple correlation coefficients using the standard tables and levels of significance is always a test of the null hypothesis that the true correlation in the population is exactly zero. Hence it is frequently observed that "significant" correlations do not always indicate a strong association. Another measure of the strength of the association is the square of the correlation coefficient, which rather roughly expresses the percent variability in the dependent variable associated with the independent variable (or variables). For example, the simple correlation between water insoluble matter and Suitability Factor is -0,1jO86 significantly different from zero at the 5-percent level. Yet, i^ X 100 « 16,7 percent of the variability in Suitability Factor is associ- ated with water insoluble matter leaving 83,3 percent still to be accounted for. The value of r^ or (r2) therefore should always be kept in mind when interpreting the significance of correlation coefficients. Analysis of Variance The analysis of variance was applied to the information on history of samples, to determine what affect these factors had on sample variability. The technique used was that of a single classification. The data were not extensive enough to allow a cross-classification of areas, processes, and dates. No attempt was made to evaluate the factor plant because so many plants fere represented by only one sample. These data are tabulated in tables 6 to 9, - 3 - DISCUSSION Simple Correlations Related to the Suitability Factor The analyses show that there was less correlation than had been supposed originally between the Suitability Factor and the 10 charac- teristics listed in table 1, Rather unexpectedly, Uie total ash had the highest simple correlation coefficient, namely r « .60"*, signifi- cant at the 0,1-percent level. Ammonia was the only other character- istic which was positively correlated, namely r " ,U0, significant at the 5-percent level, A set of values not charted showed the greatest negative correlation} namely, corrected protein with a value of r •= -."^S, indicating that, when the corrected protein (crude protein corrected for ammonia nitrogen) is high, the solubles usually gave a good dry product. Only one other characteristic, water insoluble matter, had a negative correlation, significant at the 5-percent level. These four characteii-stics, that is, a sample having low values for total ash and ammonia, and high values for corrected protein and water- insoluble matter, when taken together, indicate that it should dry on a drum dryer to give a satisfactory dry product. It might be noted that the very high simple correlation coefficient of total ash to specific gravity (r » ,75), which is developed in the first of the multiple cor- relation studies, results in the rather unusual relationship of specific gravity to the Suitability Factor (r « ,32), This value just misses being significant at the 5-percent level. There was almost no relation- ship whatever of dry solids, fat, pH, refractive index, or viscosity to predict whether a sample of condensed solubles will make a satisfactory dry product, (r2 x 100 ■ 0 to U percent). Minor Simple Coirelations Other simple correlations showed, as expected, a highly significant relationship of fat to separation of oil during drying (r " ,82) and an almost as positive relationship of ammonia to the production of objec- tionable odors noted during the drying operation (r ■ ,62), As was pointed out on page 2k of the Research Report, high values for anmonia resulted from more or less deliberate spoilage of the stickwater before evaporation, which is sometimes used as a means of reducing viscosity of the finished condensed solubles. Ammonia content of the condensed solubles had a barely significant positive correlation to the rather annoying hot plasticity of the dry solubles on. the drums, often observed d\iring the drying process. » Note all values of r have been rounded to the nearest hundredth. - U - Table le —Simple correlation values Factor correlated r r2 X 100 Suitability Factor tot •^ol779 O.I4OO8* -0,5595*** 0,0922 0.5975*** -0,>4086» 0,0156 0,1683 0,3208 -0,2039 iWcent 3 16 31 1 36 17 0 3 10 h Dry solids Ammonia Corrected proteia Fat Total ash Water insoluble matter pH Refractive index Specific gravity Viscosity Fat tot 0,8199*** 67 Oil separation AiTOonia tot 0,3711i* 0,6177*** 38 Hot plasticity Vapor odor * significant at the 5-percent level; **, at the 1-percent level| and »♦*, at the 0,1-percent level. - 5 - Multiple Correlations Several very interesting nmltiple correlation relationships were found o Factors Determining Specific Gravity The data in table 2 confirm the conclusions stated in the section on Specific Gravity on page 27 of the Research Report, that specific gravity is affected by fat content and total ash. Total ash was positively cor- related and fat negatively correlated to the specific gravity to an extent that the combined value for R " ,93, or R^ accounted for 86 percent of the variability. This value is high enough that formulas could be derived predicting the third value based on any other two, if any practical purpose was to be gained by such calculations. Factors Affecting Total Dry Solids The conclusion reached from inspection of figure 13 and stated on page 2k of the Research Report, namely that there was little correlation between the dry solids content and either the specific gravity or refrac- tive index, was also confirmed by the statistical analysis, table 3« There was no significant simple correlation of either one to the dry- solids content, and the combination gave no appreciable improvement. The R^ value was only lit percent. Factors Affecting Viscosity The next study was suggested in the hope that differences in composition might explain the very great range in viscosity of the condensed solubles that had been observed in the samples tested. The conclusion reached fix)m the data in table h was that none of the variables tested, that is, dry solids, ammonia, fat, total ash, or water insoluble matter, either indi- vidually or collectively could be significantly associated with variability in viscosity. Factors Affecting Moisture Absorption In a similar manner, four variables were selected for study to determine their effect on the moisture absorption of the dry solubles. The data in table 5 show that individually, none of them, ammonia, corrected protein, fat, or total ash had a significant simple correlation coefficient, but it was suggested that the improvement considering all variables is quite good. The mxiltiple correlation just lacked significance at the 5-percent level, and consequently, it is not of great practical value in predicting the probability that a given sanqjle of condensed solubles would yield dry solubles having low moisture absorption. - 6 - Table 2«— -jftiltiple correlation with specific gravity Factor correlated Fat Total ash Specific gravity Dry solids Fat Total ash 0.6890«*» -0.2618 -0.13U9 -0.2790 -0,5U6l*» 0.7535*** Notet R - Oo9282»»» with 28 d.f . and R^ x 100 - 86 percent. Total ash alone will account for about 57 percent, but the addition of dry solids and fat increases this to 86 percent. - 7 - Table 3.— Multiple correlation with dry solids Factor correlated Specific gravity Dry solids Refractive index Specific gravity- 0.03U7 0.2101; -0.2790 Notet R - 0.3755 vith 29 d.f . and R^ x 100 ■ 2k percent. Including both refractive 5jidex and specific gravity results in no ap- preciable improvemento - 8 - Table U»— Multiple correlation with viscosity aaiBKa^^Ka^^s Total Water ■*^a«s«aB Factor correlated Annnonia Fat ja^H insoluble Viscosity matter Dry solids -0.632ii*** 0^890*** -0.2618 0.31;95» 0.1352 Ammonia -0.ii563 processes^ and dates of collection Variable Area means^ Test i. 2 3 h 5 6 Ifuinber of samples 1 k Hi k U 5 Dry solids 52.5 Ii9.8 U9.2 U3.8 U5.3 Uh.9 ■»* Ammonia 0.39 0.57 0.93 1.3U 1.12 2.72 ** Fat 8.3 7.2 8.1; 2.8 3.0 S.5 ■Si* Total ash 7.7 9o2 9.3 10.6 6ol 10.6 *♦ Water insoluble 3.8 5.3 Uoil 3.7 k.^ 5.0 N.S. matter Viscosity 775 829 70 137 70 167 N.S. (Omitting 3000° 775 106 70 137 70 167 »* value) ♦ The areas represented by the nuabers are as follows » l) Long Island, Northern New Jersey, 2) Southern New Jersey, Delaware, 3) Chesapeake Bay, U) North Carolina, Florida, 5) Gulf of Mexico east of Mississippi delta, 6) Gulf of Mexico west of the delta. Notet Analysis of variance ignoring processes and adjusting for the vmequal numbers in each area indicates that areas may have some effect in all cases except water insoluble natter. Note that including the one observation of viscosity equal to 3000** so increases the variability that even the large differences cannot be regarded as significant. Qnission of this observation changes the significance of values for area 2, but not for the other areas. Specific comments on data In table 6: Dry solids The values for areas U, 5, and 6 appear to differ from, those of areas 1, 2, and 3* Ammonia The values for area 6 appear to be a good bit higher. The other areas ore not different among themselves. Fat The values for areas U and 5 are significantly lower than for the others. Total ash The values for areas 1 and 5 are significantly lower than for the others. Viscosity After caiitting the value of 3000° in area 2 sanqjle the analysis indicates that the values in area 1 may be regarded as higher than the others which do not differ among themselves. - 13 - Table 7 « "^Process means Variable Mean process number | Test 1 2 3 Number of samples 23 s h Dry solids U8.2 hl.h U;.08 N.S. Ainmonia 1.10 0.65 2.61 ■»« Fat 7.1 6.5 2.7 ♦ Total ash 8o9 9.1 11.3 N.S. Water insoluble li.7 3.9 U.2 N.S. matter Viscosity 87 81i9 123 N.S. (Quitting 3000° value) 87 312 123 « Note: The test results are for processes after ignoring areas and dates. Specific comments on data in table 7l Ammonia The values for process 3 are definitely higher than for 1 and 2j and processes 1 and 2 do not differ significantly. Fat The values for process 3 are significantly lower than for 1 and 2. Viscosity The values for process 2 still appears to be the highest when the 3000° value is omitted* - Ua - Table 8«—^feans by month of saiiple collection Variable July Aug, Septe Oct. Novo Test Number of samples It 13 7 7 1 Dry solids U9.5 1*6.3 U6.5 h9.h 1;3.7 N.S. Ammonia 1.26 1.57 lol6 0.72 0.1i2 N.S. Fat 8.5 5.5 6.2 7.9 lo3 N.S. Total ash 10.1 8.7 9.7 9.1 9.6 N.S. Water insoluble U.7 U.8 U.2 U,7 1.0 N.S. matter Viscosity 88 170 98 1*90 58 N.S. (Ondtting 3000° 88 170 98 72 58 N.S. value) Notot The means do not differ significantly compared to the "within the month" variation. Through November has several means much lower thaji the others^ the fact that this is based on only one sample means that it carries very little weight in the overall analysis. - 15 - Table 9«-— Number of observations in the area-process subclasses Process Area number* 1 ^ ^ Total 1 2 3 h ^ 6 1 2 3 0 1 0 2 2 0 0 0 1 1 2 ii 0 0 2 1 2 23 u Total 1 h li; J k 5 32 » The areas represented by the numbers are as follows: 1) Long Island, Northern New Jersey, 2) Southern New Jersey, Delaware, 3) Chesapeake Bay, k) North Carolina, Florida, $) Gulf of Mexico east of Mississippi delta, 6) Gulf of Mexico west of the delta* Note: It is clear from this table that means for process 3 included in only areas h and 6, Hence the conclusions regarding the effects of process 3 are not independent of similar conclusions regarding areas h and 6» Again, areas 3 and 5 contain only process 1, hence a conQ)arison of areas 3 and 5 depends on a single process and cannot be generalized to include other pro- cesses. Some caution therefore should be exercised in claiming great generality for the results of these analyses* - 16 - SUWIART 1, The following significant simple correlations were found i a. Total ash and amnonia were positively correlated (low value for desirable effect), and corrected protein and water in- soluble matter were negatively correlated (high value for desirable effect) to the Suitability Factor, b. Fat content of condensed solubles was positively correlated to oil separation during drying, c. Ammonia content was positively correlated to the development of objectionable vapor odors during diTing, d. Ammonia content was positively correlated (but at a barely significant level) to hot plasticity of the dry solubles on the dryer drums, 2, The multiple correlation studies indicated: a. Total ash was positively correlated and fat content negatively correlated to the specific gravity of the condensed solubles. The combination of these two factors accounted for almost all the variability in specific gravity, b. Although four selected characteristic components, namely ammonia, corrected protein, fat, and total ash, were individually not significsmtly correlated to moisture absorption of the dried solubles J collectively, they almost had a significsint correla- tion at the 5-percent level, 3, An analysis of vsiriance indicated: a. The location of plant had an effect on dry solids j lower values being observed in the three Southern areas, b. Ammonia was significantly higher, and fat lower in solubles pro- duced in hot-air evaporators as compared with those produced in multistage vacuum evaporators, Co Viscosity was significantly higher in condensed solubles sampled directly from the evaporators, Uo The statistical analysis showed that there were no significant correlations in the following data: a. Neither the specific gravity nor the refractive index was cor- related to the total solids content of the condensed solubles. - 17 - b. None of five selected characteristic components of the condensed solubles, namely total solids, ammonia, fat, total ash, or water-insoluble matter, neither individuallj or collectively accounted for the large range in viscosity of the condensed solubles, c. In general, the number of observations in each subclass for area and process were too few, with several areas not having any of the process groupings, to permit any generalizations as to the significance of these data except as indicated in 3a, b, and Co d. The data were insufficient to permit determining any seasonal effect on composition of condensed solubles. - 18 - CONCLUSION The original data published in Research Report kS show that the 32 randomly collected samples of condensed solubles differed considerably in chemical and physical characteristics, and produced dry solubles of widely varying properties. No single characteristic of the condensed solubles showed a degree of correlation to the desired properties of a dry product that would justify its selection as a criterion of choice of condensed solubles for use in producing dry solubles. The statistical analyses suggest, however, that condensed solubles with low ammonia and high corrected protein content, and low ash and high water insoluble matter content would diy easily and produce a satisfactory dry product. The first two relate to the freshness of the stickwater from which the condensed solubles are produced. Specifically, an ammonia content less that 1,25 percent and total ash content less than 8,5 percent; and cor- rected protein content more than 29 percent and water insoluble matter more than haS percent would be the suggested minima and maxima. It has not been possible to test the practical aspects of this conclusion since none of the producing plants is equipped with drum drying equipment at the present time* - 19 - INT.-DUP. SEC., RASH., D.C.939I I ll I ll II I r Ml 5 WHSE 01242