1 15 4 https://www.nal.usda.gov/exhibits/speccoll/files/original/4bb27c46613704dd7a53cea895d5ee84.pdf 237206a28c86f8aa4ffbe86c9a9fd9c5 PDF Text Text Item ID Number Author 01700 Carroll, Ray J. Corporate Author ROpOrt/ArtlGlO Title Typescript: Report #3, The Effect of Sampling Battalions Rather than Individuals, August 1982 Journal/Book Title Year °ooo Montn/Oay Color Number of Images D 5 Descrlpton Notes Monday, June 11, 2001 Page 1701 of 1793 �Mi REPORT n i' THE EFFECT OF SAMPLING'BATTALIONS RATHER THAN INDIVIDUALS j '• j l.J. Carroll • JAugust 1982 In my letter of August; 14jto M.E. LeVois, I raised the possibility that sampling:] from; battalions and thehlisampling individuals could have different statis- i W 1!:V . j - ' . i . iil'i : tical properties from merely taking a simple random sample of individuals. The former method is mentioned by,UCLA in their protocol, but it is clear that they I •""«.' i ' : '! , intend tojuse the latter as a basis for analysis. I think it is important to ' :• I • .". understand the difference between the two and to investigate the effects of this difference. This report is a preliminary analysis of this difference. While the latter method is called simple random sampling (SRS), the former i method might best be called two-stage cluster sampling (TSCS). The two methods j • ' t i are illustrated in Figures tfl-and #2. j •: ' . ; 1 '! : !'' | Suppose there are a total of M battalions and, for simplicity, assume each ! : i ; - /••!.' j .i ' i ! : :ji battalionjhas m individuals. 'Let battalion #i; (i = 1,2,..., M) have disease . i '••'•!• : and ';i rate p. suppose ' ! overall ! disease i •!;; ili-! the: ' I _ • rate is p. Suppose we randomly select ; •Mi 1 ! '-f1! 1: ! J ! i j i i:. N battalions and then select n jindividuals per'.selected battalion. Then the 1 ,\ i estimated probability of disease ,is (for either method) the observed proportion of diseased individuals. If the probabilities;of disease are all :fairly small ; , ! : i i "j!|i : • (say less I than 3% in every battalion), then the variances are approximately ;! I ' • ill ' „ 'Vanance(SRS) = ^ — P • ' ' ! ' mnc". • (1-nvn/MN) — ' ; i (| .. f , i Variance(TSCS) SjiilB/ll)- + ¥ «x j.w.ji'ws* v, & vui~y — . 777 n r i ' i '-S ;' '. I'-i'l',-).^ • ! •' 1 ' ': i ; '*; : i m ' £l ! ' " * • *' ' !' i '''I -i'l :j 't^'ij j i '! ::' L M ^ ^ "1 -*1 "] . : Suppose'there are M = 500 battalions, of whichjwe sample m * 70. ^Suppose that i : ; ) ' i ji'|;ri «j|.i.j, < t i I ; j '! • ' jij' j '.. j' j|| �each battalion has N - f>00 individuals, of whom we sample 100, Pui'the?, §V»ppPse that, the disease probabilities are all less than 3%. Then, to a degree of ap.i proximation, , . Variance (SRS) = (.0118)2p i ' 'iir ' »~ Variance(TSCS) '= [.0001149 + |:0030715p]p . Table #1 compares the ratio of'these two quantities for various values of p. The exact numbers in Table j#l are not particularly critical, especially since any real sampling plan wi.ll include some stratification. Nonetheless, my calculations indicate the following: i| . . .:* . .! (i) As a general strategy, we should sample as many battalions as •i i : .'Mi • •'!• possible, with appropriate stratification, i J •! | i ' (ii) For larger sample:sizcs on the order of 6,000 per group, if the event rates are small the effect of TSCS will not be too great, I , (iii) If the event rates are large, TSCS will be significantly less i •": efficient than SRS, However, in this instance, we will still have acceptable statistical powers (see Report #2). i ., , This^report has notjaddressed certain problems, such as confounders and misi ,1 <, classification. Also, I have;assumed the event rates are fairly homogeneous ' • • ' I i • ; 'ii iii; I ;!!: across battalions; this might\be a questionable assumption, as it is conceivable : :I • j ; '! ; that a few battalions had extremely high exposure and event rates. : • '•!; ii ' ' ••' study must be guided by the' practical nature of the data set. i Further �Figure _//1 SIMPLH RANDOM SAMPLING OF TRN INDIVIDUALS FROM A TOTAL OF FOUR BATTALIONS. ,Battalion #1 Battalion 112 Battalion Battalion #4 -X X X X X X X X X X J, o oo x x Note: © © On average, we will choose someone from every battalion. tee this by taking a stratified sample. ; We could guaran- *^*1 Figure #2 TWO-STAG!- CLUSTHR SAMPLING OF TRN INDIVIDUALS, SELECTING AT RANDOM TWO OF FOUR BATTALIONS. •i,'?! : Battalion til Battalion Battalion ix x x x x O | O Dip O X X X X X o; o bio o i * * * * * •ii * *' * *, * Battalion «f4 © <& © ffl © © © e ® ® Select Battalions #2 and #3 i j j i :•' . Battalion ;';!' Battalion ( x-%.: • ; ;: ,00000 I fo o o o o I U i / / O O O O O • ; :!• ' , ! . ! : •; * * * * * i Note: As opposed to simple random or stratified random sampling, in cluster sampling there is no chance of selecting one or more individuals from i| I every battalion. t ••"i! • 1$ ! : i ii .! ' �TABLH #1 Approximate Value of P .005 ,010 .020 Variance *(TSCS) .95 1.04 1.26 I !| �Appendix, Report #3 Variance (SRS) - -^ —'/—.'„ p v( i _lp ) inn Variance (TSCS) = ' - , * in I ' nm c2 2 , M S/ = -JT V (p.-p)2 1 n-1 .'•, 1 1 ' ' M S 2 = M(¥-lT .? Pi If the event rates arc all small, — 2 ] —2 (p.-p) < j p on average. � Dublin Core The Dublin Core metadata element set is common to all Omeka records, including items, files, and collections. For more information see, http://dublincore.org/documents/dces/. Title A name given to the resource Alvin L. Young Collection on Agent Orange Description An account of the resource <p style="margin-top: -1em; line-height: 1.2em;">The Alvin L. Young Collection on Agent Orange comprises 120 linear feet and spans the late 1800s to 2005; however, the bulk of the coverage is from the 1960s to the 1980s and there are many undated items. The collection was donated to Special Collections of the National Agricultural Library in 1985 by Dr. Alvin L. Young (1942- ). Dr. Young developed the collection as he conducted extensive research on the military defoliant Agent Orange. The collection is in good condition and includes letters, memoranda, books, reports, press releases, journal and newspaper clippings, field logs and notebooks, newsletters, maps, booklets and pamphlets, photographs, memorabilia, and audiotapes of an interview with Dr. Young.</p> <p>For more about this collection, <a href="/exhibits/speccoll/exhibits/show/alvin-l--young-collection-on-a">view the Agent Orange Exhibit.</a></p> Text A resource consisting primarily of words for reading. Examples include books, letters, dissertations, poems, newspapers, articles, archives of mailing lists. Note that facsimiles or images of texts are still of the genre Text. Box The box containing the original item. 063 Folder The folder containing the original item. 1700 Series The series number of the original item. Series III Subseries III Dublin Core The Dublin Core metadata element set is common to all Omeka records, including items, files, and collections. For more information see, http://dublincore.org/documents/dces/. Creator An entity primarily responsible for making the resource Carroll, Ray .J. Title A name given to the resource Typescript: Report #3, The Effect of Sampling Battalions Rather than Individuals, August 1982 Subject The topic of the resource Agent Orange Exposure Study study protocol cohort study ao_seriesIII https://www.nal.usda.gov/exhibits/speccoll/files/original/bef3e3fd9f02a5c3a14d1c89984b77d1.pdf f5fa65935a6165dc90e58cca4ac7f456 PDF Text Text Item ID Number °1699 Author Carroll, Ray J. Corporate Author RflpOrt/ArtlOlO TltlO Typescript: Report #2, False Negatives and their Effect on Power of Statistical Tests of Differences in Disease Rates, August 1982 Journal/Book Title Year 000 ° Month/Day Color Number of Images DOSCrlptOU NOtBS D 4 See ltem 1773 for a letter ^rom Carro"that mentions this report. Monday, June 11, 2001 Page 1700 of 1793 �REPORT 112 FALSE NEGATIVES AND Till:111 EFFECT ON POWER OF STATISTICAL TF.STS OF DIFFERENCES IN DISH ASF. RATES R.J. Carroll i August 1982 In REPORT #1, especially Appendix H3, I calculated statistical power for tests i' ' ' to see if disease .rates are the-same for exposed and non-exposed soldiers. I concluded that, if the disease rate in the non-exposed population was at all substan| • ; "-« •• tial (say 5%), then samples of 6,000 for each group would give very high statistical power for detecting a relative risk due to exposure of 2.0 or more, even if 1 " there was substantial misclassifixation of exposed into the non-exposed group ,i (false negatives). In this report, I will study the effect: of false negatives on statistical! power when the disease rate in the non-exposed population is fairly small. \.i i; 'To review terminology, recall that ! : j! ' ' ;;l . i p = disease probability;for truly exposed • i !• q = disease probability;for reported exposed ~ p. ' i "' i' p '= disease probability,for truly not exposed q '= disease probability for reported not exposed .•' ••• i 1 r I ~ ep,/p- = true relative risk true l*2 . . ( r ; i ' = ^i/^o " observed i iRelative risk OUi? A »•• ' 6 =ifraction of false negatives i N •= sample size for reported exposed N? = sample size for reported not exposed. .'i Before1constructing a test for equality of exposed and not exposed, we need : ! !II two concepts. The level or Type 1 error a of a test is the probability (usually a ; ,• : i I 1 .05 or*! 5%) that we will ' i conclude that theI exposed have different disease proi . ! '1 i ' in babilitiesi than do the.non-exposed, but in reality the two are not different. �The power of a test (V - 3) is the probability we will conclude that the exposed differ from the non-exposed when they really do differ. Generally, we want small Type I error but high power; UCLA chooses a = .01, 1-0 = .95. One point not made clear in the UCLA protocol is that they compute sample size based upon a one-sided test and not the usual and more conservative twoi ; In this particular instance 1 think1what they have done is justifi- sided test. ! ' able; we can hardly expect exposure to agent orange to have improved health status, If we assume equal sample sizes so that N I = N? = N, then based upon UCLA's arcsin transformation, the statistical power of the usual test is given by the formula 1 - <I»(z.. J. ~0t where z - /-=-(2arcsin/ii v ^ •*• - 2arcsin/q.J) £ is the (l-a)100% pcrccntile of a standard normal distribution function J. ~*vJt ' $ and q2 = 6p + (1-<S)P2 (Report #1, Appendix If we want to detect a true relative risk r of at least r. ., we will then : . true detect p . I • 'i need a minimum sample size of at least i M N = : 2(z if we want power of 1-3. Note that this is exactly the formula on page 108 of the UCLA protocal as long as there are no false negatives. Some idea of the effect of i false negatives can be learned from Table #1. Figure #1 shows this effect even i , . more dramatically. Note that if we desire a Type I error of a = .01 and power i !' 1-3 a .95|to detect a true relative risk of 2.0 or more, then we need about 4,500 soldiers inieach group if there are no false negatives; we will need over 7,600 �soldiers in each group if there are 20'-i false negatives. This report shows that for diseases which are fairly rare, the effect of misclassification can be severe. EFFECT OF FALSK NEGATIVES ON CALCULATION OF SAMPLE SIZE a • 1-3 r. . . detect lp~ 2 6 UCLA N Actua 9,124 11,721 15,397 20,828 29,305 4,528 5,815 7,637 10,328 14,527 2 , 230 2,863 3,757 5,078 7,138 .01 .95 2.0 .01 0.0 .1 .2 .3 .4 9,124 9,124 9,1 24 9,124 9,124 4,528 4,528 4,528 4,528 4,528 .01 .95 2.0 .02 0.0 .1 .2 .3 .4 2,230 2,230 2,230 : 2,230 2,230 .01 .95 2.0 .005 0.0 .1 .2 .3 .4 N__ �«4 l or \ \ f r \ u \ --f-O! � Dublin Core The Dublin Core metadata element set is common to all Omeka records, including items, files, and collections. For more information see, http://dublincore.org/documents/dces/. Title A name given to the resource Alvin L. Young Collection on Agent Orange Description An account of the resource <p style="margin-top: -1em; line-height: 1.2em;">The Alvin L. Young Collection on Agent Orange comprises 120 linear feet and spans the late 1800s to 2005; however, the bulk of the coverage is from the 1960s to the 1980s and there are many undated items. The collection was donated to Special Collections of the National Agricultural Library in 1985 by Dr. Alvin L. Young (1942- ). Dr. Young developed the collection as he conducted extensive research on the military defoliant Agent Orange. The collection is in good condition and includes letters, memoranda, books, reports, press releases, journal and newspaper clippings, field logs and notebooks, newsletters, maps, booklets and pamphlets, photographs, memorabilia, and audiotapes of an interview with Dr. Young.</p> <p>For more about this collection, <a href="/exhibits/speccoll/exhibits/show/alvin-l--young-collection-on-a">view the Agent Orange Exhibit.</a></p> Text A resource consisting primarily of words for reading. Examples include books, letters, dissertations, poems, newspapers, articles, archives of mailing lists. Note that facsimiles or images of texts are still of the genre Text. Box The box containing the original item. 063 Folder The folder containing the original item. 1699 Series The series number of the original item. Series III Subseries III Dublin Core The Dublin Core metadata element set is common to all Omeka records, including items, files, and collections. For more information see, http://dublincore.org/documents/dces/. Creator An entity primarily responsible for making the resource Carroll, Ray .J. Title A name given to the resource Typescript: Report #2, False Negatives and their Effect on Power of Statistical Tests of Differences in Disease Rates, August 1982 Subject The topic of the resource Agent Orange Exposure Study study protocol cohort study ao_seriesIII https://www.nal.usda.gov/exhibits/speccoll/files/original/663962cc1aeef0bc6f57ec5330fbf986.pdf 5f808b975326087d406bdc4a89072489 PDF Text Text Item ID Number 01698 AllthOT Carroll, Ray J. Corporate Author RepOrt/ArtldB TltlQ Typescript: Report #1 (with typographical corrections), False Negatives and their Effect on Estimates of the Risk of Exposure to Agent Orange, August 1982 Journal/Book Titlo Year 000 ° Month/Day Color Number of Images D 9 Doscripton Notes Monday, June 11,2001 Page 1699 of 1793 �IMPORT #1 (wil.li typographical corrections) FALSE NEGATIVES AND THEIR EFFECT ON ESTIMATES OF Till- RISK OF EXPOSURE TO AGF.NT ORANGE R.J. Carroll August 1982 I am going to assume that a large random sample is taken from individuals known to be at risk (e.g. combat troops) and that another large random sample is taken from those thought not to he at risk (rear-echelon troops)'. As explained to me, this is not really possible because only battalions and not individuals can easily be sampled for their exposure levels. I will also assume that all those reported to have been exposed actually were exposed: no false positives. A certain fraction 6 are false negatives, i.e., if <5 ~ .25, 25% of those thought not; to be at: risk (exposed) actually were. These false negatives do have an effect on estimating risk of disease due to exposure. Let p. = true probability of disease for exposed individuals, and let p« = true probability of disease for non-exposed individuals. The relative risk is defined to be r true = VP2 ' In practice we night get the following table: Reported Exposed "sick" "well" N N E N N r n N, 1 The sample sizes N Reported Not Exposed N NE N 2" NE N, 2 and N? of those reported exposed and reported not exposed (re- spectively) have been fixed in advance. N.. represents the number of "sick" indiLi viduals among those reported exposed, while N .,, is the number of "sick" individuals �among those reported not exposed. The observed relative risk can be computed by A i> (N,/N ) __ „. ._.,* _____ observed ~ (NIN ti /N«) ^ = Because of i the false negatives, the observed relative risk estimates not the true ! value r but rather true observed „_____. ____ <5 rtrue + (l-o) For example, if the true relative risk is 2.0 (exposed are twice as likely to be "sick" as non-exposed) and if we have a false negative rate of '25%, then we would report a relative risk of only A ^ Observed * ( ' 2. ' = 1<6 • Another way to look at this example is as follows. Suppose 10% of the truly .i exposed become "sick." Since the true relative risk is 2.0, only 5% = 10% * 2.0 of the truly non-exposed become "sick." However, because of the false negatives, we will announce that 10% f 1.6,= 6.25% of the reported non-exposed become "sick." TABLE J (Selected Values) True R e 1 at i v e JR. is k _ {lalse__Negati.ve_ % _ Observed Rej.ati.ye Risk ,: 1.0 0 1.0 •' 1.0 ;! 50 1.0 1 2.0 , 0 2.0 ,; ' 10% 1.82 : 25% 1.60 50% 1.33 3.0 0 3.0 10% 2.5 25% 2.0 50% 1. 5 4.0 i ! i ; 0 10% 25% 50% 4.0 3.08 2 . 29 1.60 �THE EFFECT A/EGA r s V f i S .. _ _ ^) v _ 70' i.i - ,:j. U i • :•!! OF OM 30%- �A l ' I M - N O I X "1 (8/14/82) p, ~ P (Sick/Exposed) p- - P (Sick/Not Exposed) True relative risk of exposure is ' ' r truc :i - P (Sick/Reported Exposed) 1 r i i q = P (Sick/Reported Not Rxposcd) q fc» J- ! i Suppose there arc n6"' false positives, i.e., . i '• : I Pr (Exposed/Reported Exposed) = 1 . Further, suppose a certain percentage of false negatives is possible, Pr (Exposed/Reported Not Exposed) = 6. Then, the probabilities q 1 ,q ? solve 1 I «£ q2 = Pr(Sick/Hxposcd, Report Not Exposed)6 + Pr(S.ick/Nbt Exposed, Report Not Exposed) (1-6) . If we fui'ther assume that whether the person becomes sick depends only on exposure and not on reported exposure, we get c l<-> Z = ^PiI * (l-^)P-i£. Thus, by using the misclassified table, you will be estimating an observed relative risk of exposure as obs q q? p 3!p.+(l-6)p- (Pi/Pi) [S'l)-*)! 'i = * /(6r * (1-6)} '. true true > i! r �In this instance, the observed relative risk will underestimate the true relative risk, and this bias depends heavily on the false positive rate <S. What about the excess number of cases? Per 1,000 individuals, 1,000 p truly exposed become sick 1,000 p truly not exposed become sick True excess number/1,000 is 1,000 (prp2) = 1,000 (rtrue-l)p2 . Thus if 1% of the trtfly not exposed become sick and r "C3TLIG = 2.0, then from 1,000 truly exposed individuals we can expect 10 more sick persons than we can from 1,000 truly not exposed. However, if we have a false negative rate of .50 (50%) , then our estimated excess number will not be 10 but will, be 1 0 0 Cr 0 ' observed~1)p2 * 1.000(1.33-1)(.01) = 3.3 . (see Table 2) This is a rather dramatic difference. A technical note. My definition of false negative is as on page 4 of Fleiss, not as on his page 135 (Section 11.2). His calculations in Section 11.2 are for s a retrospective study (it of sicks and wells fixed in advance) . TABLE 2 (More Details than Table 1) True Relative Risk False Negative Rate (Not %) 0.000 0.025 0,050 0.075 0.100 0. 125 0.150 0.175 0. 200 0.225 0.2*0 0.275 0.3CO O.J25 Observed Relative Risk .00 .00 .00 .00 .00 .CO .00 .CG .00 .00 .00 .00 .00 .CO �TABU: 2 (More D e t a i l s t h a n Table 1) True R e l a t i ve Risk 1 i l i 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 '* 4 '4 14 <4 4 1 <i 4 4 4 4 4 <i D 4 <l 4 4 4 4 False Ncgati ve Rate [Not Hi) O..J50 0.315 0.400 0.425 0.450 0.475 0.5CC O.OCO 0.025 0.05C 0.075 0.100 0. 1i5 0. 150 0.175 0.200 0.225 0.2SC 0.275 C.3CC 0.315 0.350 0.375 0.400 0.425 0.450 0.475 0.5CO O.OOU 0.025 0.05C 0.075 0.1CO 0. 115 0.150 0. 175 0.200 0.225 0.250 0.275 0.3 CO 0.315 0.350 0.375 0.400 0.425 0.4SO 0.475 0.5CO 0.000 0.025 0.05C 0.075 C.1CC o. i;s 0.150 0. 17!; 0.200 0.225 0.250 0.275 0.3CC 0.325 C.35C 0.375 0.400 0.425 0.450 0.475 0.500 Observed Relative Risk 1.00 1.00 1.00 1.00 1.CO 1,00 1.CO 2. CO 1.95 1.90 1.86 1.81 1.77 1.73 1.70 1.66 1.63 1.6C 1.56 1.53 1.50 1.48 1.45 1.42 1.40 1.37 1.35 1.33 3.00 2.85 2.72 2,60 2.50 2.40 2.30 2.22 2.14 2.06 2. CO 1.93 1.87 1.81 1.76 1.71 1.66 1.62 1.57 1.53 1.50 4.00 3.72 3.47 3.26 3.07 2.90 2,75 2.62 2.50 2.38 2.28 2.19 2.10 2.02 1.95 1.68 1.81 1.75 1.70 1.64 i.eo cont'd �API'liNUIX 112 J8/J4/_82) Suppose we sample N posed. who arc reported exposed and N? wlio are reported unex- What, is the probability ol" detecting a relative risk different from 1 if we ignore the effects of misclassification? For this prospective study, the usual test says (assuming a fairly larj'c percentage (7.5°d) are diseased) says that the that the relative risk differs s i g n i f i c a n t l y from 1 i r ,>. r. 1 > . — r.--.....* —T 'e olis - / I = 1.96 o(q1,q2,N1,N2) . If the normal probability function is called <I>, the statistical power is - $(1.96 + Now, lop, c r obs = loy e r true - log {6r +(1-6")} e true If N I = N , then 2 0 (q 1 q 2 , N l f N 2 ) = ^-[(1-pj) + 1 - fipj - (l-i$)p 2 ] = 1[2 - p 1 O - fip - (1-5)P 2 1 '5 = Ti-f— - (l + <5) r true - ( 1 - ^ ) 1 N l p_ ' 2 Hence the power of the usual test depends not only on the relative risk but also on the probability of disease in the truly unexposed population. �AIM'J-NDIX_ ff3_ (8 Suppose we observed a table such as Reported Exposed N sick well Reported Not Exposed .; N VNK N NE N V NE i 1 N 2 How can we estimate the true relative risk, and how much power do we have for detecting relative risks different from one? large. Recall, I am assuming q is fairly From Appendix #1 (8.14.82), page 4 , _ true ~ TT - ~~iT"r~"~jT" L obs Since an estimate of q is «i = VNi and an estimate of q? is *2 the estimate of r = N NE/N2 ' is A A ^i *obs " *1 °2 and the estimate of rtrue is A It turns out that log and variance o r _ is almost normally distributed with mean log t.i*Lio �The power of the tost for r e l a t i v e ri.sk d i f f e r e n t from one is 2 - 'K1.96 - - *<+l'96 (Highly technical, but to keep all the details written down) I want to show that (*) is true. We have A log r h true - log r b true A - log r o b s ] obs - obs ... r obs . _ -_ 1 - o i obs . A _olKS~ "ob_s_ -—--- - x l f_ } !-_ Tr~~ � Dublin Core The Dublin Core metadata element set is common to all Omeka records, including items, files, and collections. For more information see, http://dublincore.org/documents/dces/. Title A name given to the resource Alvin L. Young Collection on Agent Orange Description An account of the resource <p style="margin-top: -1em; line-height: 1.2em;">The Alvin L. Young Collection on Agent Orange comprises 120 linear feet and spans the late 1800s to 2005; however, the bulk of the coverage is from the 1960s to the 1980s and there are many undated items. The collection was donated to Special Collections of the National Agricultural Library in 1985 by Dr. Alvin L. Young (1942- ). Dr. Young developed the collection as he conducted extensive research on the military defoliant Agent Orange. The collection is in good condition and includes letters, memoranda, books, reports, press releases, journal and newspaper clippings, field logs and notebooks, newsletters, maps, booklets and pamphlets, photographs, memorabilia, and audiotapes of an interview with Dr. Young.</p> <p>For more about this collection, <a href="/exhibits/speccoll/exhibits/show/alvin-l--young-collection-on-a">view the Agent Orange Exhibit.</a></p> Text A resource consisting primarily of words for reading. Examples include books, letters, dissertations, poems, newspapers, articles, archives of mailing lists. Note that facsimiles or images of texts are still of the genre Text. Box The box containing the original item. 063 Folder The folder containing the original item. 1698 Series The series number of the original item. Series III Subseries III Dublin Core The Dublin Core metadata element set is common to all Omeka records, including items, files, and collections. For more information see, http://dublincore.org/documents/dces/. Creator An entity primarily responsible for making the resource Carroll, Ray .J. Title A name given to the resource Typescript: Report #1 (with typographical corrections), False Negatives and their Effect on Estimates of the Risk of Exposure to Agent Orange, August 1982 Subject The topic of the resource Agent Orange Exposure Study study protocol cohort study ao_seriesIII Dublin Core The Dublin Core metadata element set is common to all Omeka records, including items, files, and collections. For more information see, http://dublincore.org/documents/dces/. Title A name given to the resource Alvin L. Young Collection on Agent Orange Description An account of the resource <p style="margin-top: -1em; line-height: 1.2em;">The Alvin L. Young Collection on Agent Orange comprises 120 linear feet and spans the late 1800s to 2005; however, the bulk of the coverage is from the 1960s to the 1980s and there are many undated items. The collection was donated to Special Collections of the National Agricultural Library in 1985 by Dr. Alvin L. Young (1942- ). Dr. Young developed the collection as he conducted extensive research on the military defoliant Agent Orange. The collection is in good condition and includes letters, memoranda, books, reports, press releases, journal and newspaper clippings, field logs and notebooks, newsletters, maps, booklets and pamphlets, photographs, memorabilia, and audiotapes of an interview with Dr. Young.</p> <p>For more about this collection, <a href="/exhibits/speccoll/exhibits/show/alvin-l--young-collection-on-a">view the Agent Orange Exhibit.</a></p> Text A resource consisting primarily of words for reading. Examples include books, letters, dissertations, poems, newspapers, articles, archives of mailing lists. Note that facsimiles or images of texts are still of the genre Text. Box The box containing the original item. 063 Folder The folder containing the original item. 1697 Series The series number of the original item. Series III Subseries III Dublin Core The Dublin Core metadata element set is common to all Omeka records, including items, files, and collections. For more information see, http://dublincore.org/documents/dces/. Creator An entity primarily responsible for making the resource Carroll, Ray .J. Title A name given to the resource Letter: from Ray J. Carroll to Maurice E. LeVois, August 14, 1982 Subject The topic of the resource Agent Orange Exposure Study study protocol cohort study ao_seriesIII