Chapter 6: Experimental Procedures:
6.1 General Study Design
The degree of variability of different parameters of neuroanatomy was derived from MRI imaging techniques. This was compared with the degree of variability found in a broad range of cognitive abilities to assess the level of association between neuroanatomical variation and behavioral variation. Pairs of female siblings were recruited in order to investigate the strength of these correlations within- and between-families.
6.2 Subject Selection
6.2.1 Who , Why, and How Many
Pairs of female siblings were recruited The use of females only will control for the effects of sex, while the comparison of siblings will lessen possible effects of between-family socioeconomic variability and allow for the analysis of separate within and between family relationships among the variables of interest. 39 pairs were recruited.
6.2.2 Inclusion/Exclusion Criteria
Subjects were non-pregnant females who had same sex siblings, not more than 6 years older or younger, who were also willing to participate in the study. Participation was limited to females for the following reasons:
1) Age of menarche represents an easy, non-invasive method for estimating the age of maturation (and therefore rate of growth),
2) Females are less frequently studied than males in medical studies,
3) Significant sex differences in brain size (Ho et al. 1980; Pakkenberg and Voigt 1964; Skullerud 1985), the high cost of MRI scanning, and the modest size of correlations found in previous research on these questions (Johnson 1991) suggest that a single-sex study will be most cost effective.
Subjects were screened for history of dyslexia, brain trauma, brain disease, or time spent in a psychiatric institution (subjects were not excluded for simply having had psychoanalysis, must not be taking any medication that might affect concentration, and must have normal hearing and vision (at least with corrective lenses).
6.2.3 Subject Recruitment
Subjects were recruited from the general community as well as from various college campuses in the Bay Area. Initial contact was via local newspaper advertisements, posted announcements on campuses, as well as verbal announcements in classes.
6.2.4 Consent Process and Documentation
Separate consent was obtained for the MRI and cognitive testing portion of the study, since they were performed at different times and at different places. Prior to both the cognitive testing and the MR exam the subject read the consent form and had the opportunity to have the study explained by the Investigators before signing. The subjects were also given a standard MRI screening form prior to the MRI exam. The protocol for this study was approved by both the U. C. Berkeley Committee for Protection of Human Subjects (project number 94-8-86) and the U.C. San Francisco Committee on Human Research (approval number H7063-08927-04). Copies of the approved consent forms are included in appendix D.
6.2.5 Procedures
6.2.5.1 Data Collection
Study Sites and Exam/Testing Times
The MRI exams were performed at the UCSF Magnetic Resonance Science Center (MRSC). Each MRI exam was completed within 45 minutes (scanning time itself was ~30 minutes). The cognitive testing were performed at Dr. William Wang's laboratory, the Project on Linguistic Analysis (POLA) located at 2222 Piedmont Ave., U. C. Berkeley, as well as at a private practice office in San Francisco. Subjects were given a choice of testing locations to ease their travel time. The cognitive testing portion took approximately 2 3/4 hours total. Only one subject was tested at a time.
MRI Exam Procedures
The subjects were greeted by the MRSC receptionists, given the consent form, MRI screening form, and Experimental Subjects Bill of Rights. I discussed the exam and answered any questions prior to the subject signing the consent form. A copy of the consent form was given to the subject upon request. The subject then changed into a hospital gown. Subjects were weighed, and had triceps skinfold measurements and arm circumference taken (following Harrison et al. 1988). The subject lay on the table, was positioned as comfortably as possible and then slid into the machine. Verbal contact between the MR operators and the subject was maintained continually via an intercom system. If the subject wished to end the study at any time, this was done (one subject was unwilling to continue because of unforeseen claustrophobia).
The images in this study were obtained with the following MR scanning sequence protocol:
1) Sagittal T1 scout (TE 12 ms, TR 500 ms, 26 cm FOV, 256 X 128 pixels, 1 NEX, 10 mm thick, with 2.5 mm gaps between slices). Scanning time: 2 minutes. This was used to properly align the following scans.
2) Axial T2-weighted scan with 4 echoes (TE 25 ms, 50 ms, 75 ms, 100 ms, TR 2500 ms, 24 cm X 18 cm FOV mapping onto 256 X 192 pixels, .75 nex, 3 mm thick with no gap between slices). Scanning time: 15 minutes.
3) Axial 3-D SPGR (spoiled-gradient) sequence (TE 8 ms, TR 32 ms, 45¡ flip angle, 24 cm X 18 cm FOV mapping onto 256 X 192 pixels, 0.75 nex, 1.5 mm thick slices with no gaps) (Budinger et al. 1991). Scanning time: 9 minutes.
One subject was not able to complete the MRI exam because of claustrophobia. Thus this sibling pair could not be used in the analysis that follows. In addition, two subjects were found to have abnormalities on their MRI scans. One had an arachnoid cyst exhibiting slight displacement of the surrounding cortical tissue. The other had an enlarged pituitary. Both subjects scored within the normal range on the cognitive tests. Since the question here is whether differences in brain anatomy, however caused, are associated with cognitive variability, it is not unreasonable to include them in the analyses. However, it is possible that the cause of the abnormality also independently causes cognitive deficits, in which case including these subjects might bias the results in some way. For the sake of simplicity, these two subjects (and their sisters) were excluded in the analyses below. The total number of sibling pairs used was therefore 36.
Brain Anatomy Quantification.
For the present study, only the Axial 3-D SPGR images were used because of their superior grey-white-CSF differentiation. These images were processed as follows. First, each volume image was stripped of identifying information and coded with a random number by another technician so as to ensure that the quantification of brain dimensions was done blind with respect to a given subject's scores on the various cognitive tests. Then non-brain tissue was removed from each of 124 slices through a thresholding technique available in the 2-D segmentation module of the VIDAª image analysis software package (Hoffman et al. 1992). This module automatically selects all areas of image above a given image intensity value that are contiguous with a manually "seeded" pixel, and turns all other pixels to zero. The optimum intensity value for a given scan was selected by visual inspection for slices occurring 30, 60, and 90 mm from the top of the brain. The average of these values was used to start with, but this was adjusted when necessary (some scans exhibited changes in average intensity at the lowest levels). Care was taken to ensure that all non-contiguous portions of brain tissue on a given slice were seeded. Any remaining non-neural tissue was then manually deleted. The following arbitrary conventions were used to ensure replicability: 1) the brain stem was included down to the lowest slice which also included cerebellar tissue, 2) the optic nerve was removed only were it was separable from the brain stem on at least one side (this meant that about half the optic tract and all the optic chiasm and optic nerve were removed), 3) the mammilary bodies were not removed, 4) other cranial nerves were removed from the point at which they where clearly separate from the brain stem, 5) the choroid plexus in the lateral ventricles was not removed, though if it was not contiguous with the ventricle walls it was not specifically added. These conventions only affect very small portions of the brain. The only non-zero pixels remaining in the resulting files were therefore exclusively brain tissue. Brain volume (BRAINVOL) was quantified by counting all non-zero pixels and multiplying by their size (0.9375 mm x 0.9375 mm) and the thickness of the slices (1.5 mm). Since the slices were contiguous, this resulted in complete coverage of the entire brain, resulting in very accurate estimates of brain volume.
This measure was highly reliable. Seven subject scans were selected for re-quantification specifically because they were judged by eye to provide the greatest potential for error. The RMS error for BRAINVOL on these replications was less than 3%, and the re-quantified values correlated with the original values greater than r=.95 (p<0.001). Note that these are conservative estimates because they were based on biased set of scans (i.e., ones that were thought to contain the greatest error).
Another file was then created (from the file used to calculate BRAINVOL) for which the brain stem and cerebellum (myelencephalon, metencephalon, and mesencephalon) were removed, resulting in the cerebral cortex and deep cortical structures only (prosencephalon; sometimes referred to as the cerebrum). The brain stem was removed starting at the highest level in which the internal capsule was no longer contiguous with the cerebral peduncles. On most slices there is a clear distinction between either the cerebellum or brain stem and the cortex or deep cortical structures. Questionable portions were checked on immediately adjacent slices to ensure the most accurate segmentation. The volume of the prosencephalon (PROSVOL) was quantified in the same manner as with overall brain size. The volume of the cerebellum + brain stem, reflecting the myelencephalon, metencephalon, and mesencephalon (MMMVOL) was obtained by subtraction.
Prosencephalon grey and white matter volumes (GREYVOL, WHITEVOL) were estimated by using histograms of the pixel intensities, which indicated two distinct peaks (representing grey and white matter regions). Two gaussian curves were fitted by an iterative process to the histograms, using a program written by Sundar Amartur at the Center for Functional Imaging, Lawrence Berkeley National Laboratory. Figure 6.1 shows histograms (technically, probability density functions) of the pixel intensities from three subjects, along with curves representing the sum of the best-fit gaussian curves. Comparing the two indicates that this iterative estimation process leads to a close fit with the actual data. The means and standard deviations of these best-fit curves were used to estimate the total volume under each curve, resulting in objective estimates of GREYVOL and WHITEVOL regions.
A measures of how distinct the grey and white matter regions were in the prosencephalon were also extracted from the best-fit grey and white matter gaussians. This was calculated as:
EFFECT SIZE =
Where:
meanwhite = mean of the white matter gaussian pixel values
meangrey = mean of the grey matter gaussian pixel values
swhite = standard deviation of the white matter gaussian pixel values
sgrey = standard deviation of the grey matter gaussian pixel values
"Figure 6.1 Gaussian curve fitting on pixel intensity probability density functions" \l 1 Figure 6.1: Examples of the pixel intensity probability density functions (PDF's) with fitted gaussian curves. The actual data points used to estimate the gaussian curves are shown as crosses (+). The sum of the two best-fit gaussian curves are indicated by the lines. The top two data sets were typical of the scans in this study, the bottom one was unique.
This is simply the standardized difference between the two curves (also known as the "effect size").
Cognitive Testing Procedures
During this session, several tests, a questionnaire, and a handedness inventory were given, and some anthropometric measurements were made. The order in which these were done was the same for all subjects (see the end of this section for the specific order). The details of these are as follows (the names of variables used in the analysis to follow are indicated by capital letters, e.g.: RAVEN).
1) Questionnaire
The questionnaire was divided into two parts: A and B. Part A was given in the beginning of the testing session, and Part B was given at the end. Part A asked for the following information (page 1 of the questionnaire; see appendix B) :
a) Verification of name through some form of photo ID (to be sure the sisters were related),
b) Background screening information, to double check the answers the subjects had previously given over the phone. The questions asked if the subject had any history of dyslexia, mental disease, unexplained blackouts, brain trauma, surgery in which something metallic might have been implanted (potentially screening them from MRI scanning), pregnancy, claustrophobia, hearing and/or vision problems, and use of medication.
c) Date and place of birth,
d) Number of siblings
e)Subject's current occupation.
f) Mother's occupation (both at present and while the subject was in High School)
g) Father's occupation (both at present and while the subject was in High School)
h) Years of schooling (SCHOOL-S), college attended (if any), and highest degree obtained,
i) Years of schooling of parents (mother: SCHOOL-M, father: SCHOOL-F)
j) Existence of left-handed family members (both immediate family as well as more distant relatives).
Part B of the questionnaire (pages 2-3; see appendix B for details) included the following:
a) An updated version of the Home Index of status items (Gough 1971), asking things about the subject which would suggest differences in socio-economic status (this was factored into an estimate of SES as described in chapter 8). Several of the questions asked the subject to respond both about their present situation, as well as the situation while they were living at home during the first two years of high school. The questions included (but were not limited to) the number of books, color televisions, cars, and phones the subject owned, whether they had any servants such as a maid or cook, whether they regularly go on vacation, own a computer, piano, compact disk, whether they belong to a club that requires the payment of dues, whether they have ever had any private lessons in music, dancing, art, etc. outside of school, and other questions (see appendix B for the complete list).
b) Age of menarche (MENARCHE),
c) Which of the two sisters participating got her period first (if known),
d) Date of the beginning of the subject's last menstrual cycle,
e) Total number of people they have in their address book (ADDRESS BOOK TOTAL),
f) Total number of relatives in their address book (ADDRESS BOOK RELATIVES),
g) Number of different people seen socially in the average month (PEOPLE SEEN SOCIALLY). Subjects were told to include any social activity in this assessment, ("For example, having someone over for dinner, going out to lunch, movies, etc. Anything were the point is to interact in some way with another person").
h) Number of different people talked to for more than five minutes in the last week (PEOPLE TALKED TO). Subjects were told that this could include anyone at all, even people they did not know personally, such as a bank teller, clerk, customer at work, etc.
i) Whether the subject consider herself to be "more social" or "less social" than her (participating) sister, and to rank their degree of sociality on a scale from 1 ("A social hermit; someone who does not need or seek out any social contact") to 5 ("A social butterfly; someone who loves to spend every possible moment with other people") (SELF RATING)
j) The subject's judgment of her participating sister's degree of sociality, on the same scale as above. This was included in the participating sister's data as that sister's RATING BY SISTER variable.
k) Amount of physical activity per week (hours per week),
l) Favorite physical activities (in order of preference),
m) Ethnicity
Handedness inventory and task
Each subject was also given a standard handedness inventory (Oldfield 1971), and a pencil-and-paper test of hand laterality (following Tapley and Bryden 1985). On this test, subjects put a dot in the center of as many printed circles as they can in 20 seconds. Both hands are tested twice, in the following order: Dominant hand, non-dominant hand, non-dominant hand, dominant hand. The score for each hand was the total circles completed on both trials on that hand (completed circles were ones in which more ink was visible inside the circle than outside, with borderline cases being counted as completed). The extracted measures were as follows:
HANDPERF-R = number completed circles on right-hand trials
HANDPERF-L = number completed circles on left-hand trials
HANDPERF-TOTAL = (HANDPERF-R) + (HANDPERF-L)
HANDPERF-SCORE = [(HANDPERF-R) - (HANDPERF-L)] /
HANDPERF-TOTAL
This handedness data was not analyzed further for this study, but is reported here for informational purposes only
Anthropometric measurements.
The subjects' height (BODY-HT) and sitting height (SIT-HT) were measured (following Martin et al. 1988) using a standard field anthropometer. Leg length (LEG-HT) was estimated by subtracting SIT-HT from BODY-HT. The average of the three closest triceps skinfold measurements (TRICEPS) was used to estimate body density using age/sex appropriate equations in Durnin and Womersley (1974), except for subjects of asian descent (self described) for which the female equation in Nagamine and Suzuki (1964) was used instead. Percent fat (% FAT) was calculated from density using Siri's (1956) equation. Lean body weight (LBW) was estimated by multiplying body weight (BODY-WT) by (1 - % FAT).
Cognitive tests.
The cognitive testing portion of this study involved the administration of several highly diverse kinds of tasks. The test battery was selected to tap abilities in the following broad cognitive domains: General ability ("g"), linguistic processing ability, spatial ability, prefrontal functioning, simple reaction time, short term memory, and throwing accuracy. For each task, the subject was first verbally given a brief outline of the test instructions. A full set of written instructions was then given. When the subject confirmed that she understood the directions, the test was administered. As noted below, some of the tasks had practice trials preceding the experimental trials (i.e., RAVEN, SRT, OBJECT-ID, SYNTAX, MRT, THROW; see descriptions below). Subjects were seated at a desk for all tasks (except for the THROW test).
All the computerized tests described below were programmed using Hypercard 2.1 for the Macintosh. For the SRT, OBJECT-ID, SYNTAX, and MRT reaction time tests, a special module was written which allowed the following procedure to be used: 1) the subject presses and holds the <spacebar> down to indicate she is ready of the next trial item, 2) the item is displayed on the screen after a short delay, 3) the subject releases the <spacebar> when she is ready to respond (this varies with the test; see below) and presses a specified key. The Hypercard module then returns the following information: A) The time between the screen change (when the stimulus is given) and the release of the <spacebar> ("decision time"), B) the time between the release of the <spacebar> and the pressing of another key ("movement time"), and C) the specific key pressed by the subject (allowing the program to indicate to the subject whether or not they pressed the appropriate key). The separation of the reaction time into decision and movement portions allows for a theoretically more accurate estimate of mental processes, as distinct from purely motor functions. The specific keys used by these tests were "9" and "4" keys above the letter keys (i.e., not on the keypad). The "9" was covered with a green sticker with the letter "S" (for "same") on it, and the "4" was covered with a red sticker with the letter "D" (for "different") on it. The keys immediately surrounding these target keys were covered with blank white stickers to cut down on possible distractions and to create a visual "bulls-eye" effect. For the choice reaction time tests (i.e., OBJECT-ID, SYNTAX, and MRT) the subject is instructed one each test to press the green "S" key if the stimuli presented are the same, or the red "D" key if they are different. On the standard Mac Plus keyboard these keys are about 9 cm from the center of the <spacebar> (which is where subjects placed their fingers for the choice reaction time tests). For the SRT only the green "S" key was used, which is only 7.5 cm from the <spacebar> at the shortest point.
The tests were self paced, in that the subject was not timed in between each trial: Timing began only after the subject pressed and held the <spacebar> down. Information on each trial was saved for later analysis. Programs were written which calculated the median reaction times, the standard deviations of the reaction times (as a measure of the variability of the subject's response times), and the number correct. These summary statistics were calculated only on those trials for which the subject pressed the correct key for that trial. Trials for which the subject responded faster than 150 milliseconds were excluded on the grounds that this is faster than humanly possible (Jensen, personal communication), such that the subject must have been anticipating rather than responding (this was a very rare occurrence). Trials for which subjects exhibited unreasonably long movement times were also excluded, in order to reduce the probability that a subject would change her mind after releasing the <spacebar>, thereby making some of their decision-making time appear as movement time. The reaction time programs initially gave the subject an error if their movement times were longer than 500 milliseconds, which was more than adequate for most subjects. However, if the subject consistently had movement times slower than this, the program adjusted the allowable movement times upward, so that these subjects would not go through the whole testing session getting errors for responding too slowly. The median times, standard deviations, and N's were also calculated for split halves (e.g., odd/even, first-half/last-half, etc.) in order to estimate test reliabilities. Medians were used instead of means so as to decrease the effect of extreme outliers on the subject's score. Because the individual trials were relatively short, temporary loss of concentration on a given trial could have a large affect on the mean reaction time. The use of medians instead of means therefore helps to separate out variability due solely to the ability to maintain concentration, assuming this is a distinct ability from the specific task at hand (which is admittedly an open question).
Descriptions of the specific tests administered are as follows. Note that the tests are grouped by general cognitive categories for convenience only. A number of tests have components which obviously place them in more than one of these categories. For example, the verbal fluency task is included in the "prefrontal" group of tests because it shows specificity to prefrontal damage, but it is also clearly a verbal/linguistic task as well. As was discussed above, no single test is ever a completely pure measure of a single cognitive function, and this provides the justification for the multivariate statistical techniques discussed below.
General ability ("g") task:
(a) RAVEN. Raven's Progressive Matrices (Raven 1948): This is a standard non-verbal test of reasoning ability. It is considered to be one of the most "culture-free" reasoning tests, in that it simply requires the subject to extract a pattern implied by a set of geometric pictures. The patterns get increasingly complicated, but they do not contain any obvious verbal component. For this study, subjects were given 20 minutes to complete as many matrices as they could from a sequence that included items from both the Raven Standard Progressive Matrices (SPM) and the Advanced Progressive Matrices (APM). This was done to avoid a possible ceiling effect on the SPM, which can be too easy for bright college students. The SPM consists of 5 sets (lettered A-E) of 12 items each, with each set starting with easy items which increase in difficulty. Set A is too easy for most adults, and sets B and C are only slightly harder. For this study, therefore, set A was not used and only the even numbered items were used from sets B and C. All the items from set D and E were used. This was followed by more difficult items from set II of the APM: items 27-36 (set I is generally used simply to make sure the subjects understand what they are supposed to do on the test). In order to limit the likelihood that a subject would finish all the items in less than 20 minutes, items 13-26 (set II) of the APM were placed at the end, in reverse order of difficulty (i.e.: 26,25,24...13). Only 9 subjects got to these latter questions. The actual item sequence used for this test therefore was: (SPM) B2, B4, B6, B8, B10, B12, C2, C4, C6, C8, C10, C12, D1-12, E1-12, followed by (APM, set II) 27-36, and lastly 26-13 (reverse order). The first two items were used to explain to the subject what they were supposed to be doing (see appendix B for text of the instructions given to each subject). See appendix B for a description of the instructions given to each subject for this test.
Linguistic processing tasks:
(b) VOCAB. Vocabulary portion of the Multidimensional Aptitude Battery (Jackson 1984): This is a 46 item, multiple-choice vocabulary test, requiring the subject to identify synonyms of target words.
(c) OBJECT-ID. Speed of access to information stored in long term memory. This is a reaction time test in which the subject has to indicate whether or not a picture of an object (either a square, circle, triangle or star) correctly matches a word ("STAR," "CIRCLE," "TRIANGLE," OR "STAR"). These items were chosen simply for consistency with the other reaction time tests administered for this study which use the same stimuli items. They also are extremely familiar to all normal adults, such that individual differences in the reaction times on this test are not easily explained on the basis of prior exposure to the test items. Each trial consists of one word below one picture of an object. Each word appeared with each of the objects an equivalent number of times, with the restriction that a given word was paired with its corresponding (correct) picture three times, and once each with the other three (incorrect) objects. This ensured that there were an equal number of correct and incorrect trials. Each word was therefore presented 6 different times per block, resulting in 24 trials per block (4 words X 6 trials per word). Three blocks were given for a total of 72 trials (3 X 24). Four sample trials are given, but if the subject hit the wrong key on one of them, that practice trial was repeated until they pressed the correct key. The test was entirely self paced, and a short break was offered between each block. This test followed the basic reaction time procedures described above. After a delay of 500 milliseconds the trial items are displayed, and the subject then decides whether or not the word and the object are the same or different. They are instructed to press the green "S" key if they match, or the red "D" key if they do not. For more details, including the order of words/objects presented, see appendix B.
(e) SENTENCE-VERIF and SYNTAX. Sentence verification: A computerized reaction-time task of the speed of comprehension of sentences with two types of syntactical construction. This test followed same format as the other reaction time tests (as described above). A delay of 500 milliseconds occurred between the press of the <spacebar> and the presentation of the stimulus items on this test. The subject was required to verify whether or not a given sentence correctly describe a pattern of simple geometric objects. They pressed the green "S" key if the sentence matched the pattern of objects, or the red "D" key if the sentence did not. The objects used where: black star, grey square, white circle, and striped triangle. The location of each of the four objects varied randomly across trials, with the constraint that all four object types appeared in each pattern, arranged in a square (see appendix B for examples). All four objects were used in each target picture, even though the sentences presented for verification on each trial only referred to 3 of the objects. This is necessary to ensure that no pattern (with respect to the arrangement of objects) would be associated with a particular syntactical structure. The adjectives "beside" and "above" are found in equal proportions for each set of syntactical constructions. The sentences were randomly constructed given the following constraints: A) The four object types are found in equal proportion both across all sentences as well as at each serial position in the sentence, and B) the spatial descriptors (the adjectives "beside" and "above") are found in equal proportion in the grammatical forms used. Two basic types of syntactic structure were used: right-branching (RB), e.g., "The triangle is beside the square which is above the star"; and center-embedded (CE), e.g., "The triangle which the square is beside is above the star". Most subjects find center-embedded sentences to be more difficult to process than right-branching sentences. Since this test used pairs of sentences identical in the number and type of words used, and only differed in the arrangement of words (i.e., syntax), the difference in median verification time for the two sentence types can only be due to differences in syntax processing time. The test was divided into 8 subsets. Each subset contained either RB or CE sentences, but not both, with a RB subset always followed by a CE subset. Each subset (of either kind) contained 8 sequences forms (2 preposition sequences X 4 object word sequences). Only half of the RB sentences in a given subset had the same truth value as the corresponding sentences in the following CE subset, such that the subject could not guess that a particular sentence type was likely to be either correct or incorrect solely on the basis of the trials in the subset immediately prior. Also, because there were two basic ways in which a given sentence could be wrong (i.e., in either the first or last phrase), half the time a given sentence was incorrect only with respect to its first phrase, and the other half the time it was incorrect only with respect to its last phrase. This was necessary so that subjects could not simply read the last phrase of the sentence first and respond correctly solely on the basis of this part (some subjects were found to do this in a pilot project). In order to ensure an equal number of correct and incorrect trials, this meant that each sentence type had to appear 4 times (twice correct, once only incorrect on the first phrase, once only incorrect on the last phrase). Furthermore, all sentences which were incorrect on the first phrase in the first half of the test were incorrect on the last phrase in the second half. The total number of trials was therefore 64 (eight sentence forms repeated four times for each of the two types of syntactic structures). The order of sentences were pseudo-randomized, given the constraints above, both within and between subsets with respect to sentence forms, truth value, and part of the sentence that was incorrect. Eight sample trials were given: four right-branching trials, followed by four center-embedded trials. If the subject hit the wrong key on one of them, that practice trial was repeated until they pressed the correct key. SENTENCE-VERIF was calculated as the average of the median decision times for the two types of sentences. SYNTAX was calculated as the difference between the median decision times for the two types of sentences. For more details, including the exact sentences and order presented, see appendix B.
Spatial ability task:
(g) MRT. Mental rotation (adapted from the Vandenberg and Kuse 1978 version of the Shepard and Metzler 1971 task): In this test, the subject was shown two dimensional images of three dimensional objects displayed on a computer screen. The objects are composed of individual cubes stacked and connected into a three dimensional array. Two of these images were shown side by side on screen, and the subject was instructed to decide whether or not they could be different views of the same subject, or if they must be different objects. The item pairs were in exactly the same order as those encountered on the paper-and-pencil version (assuming subjects proceed from left to right on the paper-and-pencil version). The items were individually scanned into the computer, enlarged so that they just fit into an area 6.5 cm square (all items were enlarged the same proportion from the original pencil-and-paper version), and manually retouched to maximize clarity. The objects were white, drawn with black lines, and presented on a grey background. Only the first half of the items from the pencil-and-paper version were used (40 total). The procedures of the test followed those outlined above for the other reaction time tests. A delay of 250 milliseconds was used between the press of the <spacebar> and the presentation of the stimulus items on this test. Eight sample trials were given, which were taken from those used on the pencil and paper version. The subject was allowed as much time as they wanted to view sample trials again if they answered any incorrectly. See appendix B for examples and subject instructions.
Prefrontal functioning tasks:
(h) TRAILS. Trails Test (Reitan and Wolfson 1985). This test has two parts. Part A is essentially a timed connect-the-numbers task (from 1 to 25), but the numbers are arranged in such a way that the subject can plan ahead. For example, while approaching number 19, the subject must pass close to number 20. Part B is similar, but the subject must this time alternate between numbers (1-13) and letters (A-L). These are also placed so as to facilitate planning. The instructions followed Lezak (1983). In order to correct for individual differences in speed connecting the dots together, the score was calculated as (time on Part B) minus (time on Part A). Thus the common component involved in connecting 25 dots was subtracted out of the score, leaving just the extra time needed to rapidly alternate over-learned sequences in the mind.
(i) WCST. Wisconsin Card Sorting Test (Heaton 1981): This standard test of prefrontal functioning requires the subject to figure out the sorting criteria for a set of cards without explicitly being told how. They are simply told after each trial whether their choice was correct or incorrect. There are three independent ways the cards could be sorted: By shape of the objects on the card, by number of objects on the card, or by the kind of pattern filling the objects on the card. The "correct" sorting criteria is changed (without informing the subject) when the subject demonstrates competence on a given sorting criteria by correctly sorting 10 cards in a row. This was adapted to the computer (using Hypercard 2.1 for the Macintosh) and changed to a back-and-white format for this study. The items used in this computerized adaptation were different only in the following way: red became grey, yellow became white, blue became black, green became black and white striping (see appendix B). The size that the cards appeared on the screen was approximately the same as that for the original test (the cards as they appear on the computer (Macintosh Plus) screen are 42 mm square, and the objects on them are approximately 10 mm in size). The order of presentation of the 64 cards was identical for each subject. The instructions closely followed those given by Grant and Berg (1948), with obvious adjustments required by the fact that the computer was giving the exam, not an experimenter. The test was self-paced, with the subject indicating she was ready for the next trial by moving the mouse into a box at the center of the screen. The "target" cards which the trial card was to be matched to were always located at the top of the screen in the same arrangement. The trial card appeared just below the box at the center of the screen, and the subject indicated which "target" card she wished to match it to by pressing the mouse key over one of them (see appendix B and Grant and Berg, 1948, for more details).
(d) VERBALFL. Verbal Fluency (adapted from Thurstone and Thurstone 1941): Subjects were asked to recall as many words as they could think of in one minute that start with a particular letter. Three trials were given, using the letters "F", "A", and "S". Trials were timed, and subject responses were recorded for later analysis. The instructions were given as in Lezak (1983)
(f) STROOP. Stroop test (Stroop 1935; Golden 1978): A timed test of the degree of linguistic interference on a color naming task. This test also shows specificity to prefrontal damage (Lezak 1983). This test has three parts. The first (W) measures how many words a subject can read out loud in 45 seconds. A random list of only three words are used: "Red," "blue," and "green." The second part (C) measures how many colors a subject can name in 45 seconds. A random list of red, blue, and green colored items are used (the items consist of four X's). The last part (CW) also measures how many items the subject can name the color of, but the items in this case are not X's, but are words for colors written in a contrasting color of ink (e.g., the word "Red" written in either blue or green ink). This part measures how much interference written words cause in the naming of colors. It may be though of as an indication of how much predominance linguistic processing has over color processing, with respect to verbal tasks. The instructions given the subjects followed Golden (1978). Beside the raw scores for these tasks, a composite interference score (STROOP) was calculated which adjusts for differences in reading or color naming speed. This variable is defined as:
STROOP = CW -
See Golden (1978) for justification and derivation of this formula.
Simple reaction time task:
(j) SRT. Simple Reaction Time task. This test measures the speed at which a subject can react to the appearance of something on the display screen. It uses the general reaction time format as described above. Specifically, the outline of a white circle in the center of a white background turns black after a random delay. The subject begins each trial by holding down the spacebar on the keyboard. After a random delay of between 250 and 2000 milliseconds, the trial item is displayed, and the subject then presses the green key on the keyboard (which is the "9" key above the letter keys). 32 trials were given to each subject. The same order of the same random delays were used for all subjects (see appendix B for details). As with the OBJECT-ID test, four sample trials were given, but if the subject hit the wrong key on one of them, that practice trial was repeated until they pressed the correct key.
Throwing accuracy task:
(k) THROW. Dart Throwing accuracy. The subjects threw 22 gram, brass darts with nylon shafts (manufactured by Dart Mart, Inc., in England). The target consisted of a 28 mm diameter filled black circle, marked on a 457 mm X 610 mm sheet of plain paper, which was tacked onto a corkboard. The target was located 275 cm away from the subject, 140 cm above the floor. The small holes created on the paper by the darts at the location of impact where marked by circling them with a colored pen, so that measurement could be done at a later time. Subjects where given 6 practice throws followed by 6 throws which "counted" (marked for later analysis). This whole procedure was repeated again after the last cognitive test (about one hour later), giving a total of 12 marked throws.
As mentioned above, the order in which the tests, anthropometric measurements, break, and questionnaire were administered was the same for all subjects. The specific order was as follows:
a) Questionnaire (part A)
b) Handedness inventory and test
c) Anthropometric measurements
d) RAVEN
e) VERBALFL
f) VOCAB
g) WCST (computer)
h) Break (the time varied with, and was determined by, the subject)
i) Darts (first 12 throws)
j) SRT (computer)
k) OBJECT-ID (computer)
l) SYNTAX (computer)
m) Trails test
n) STROOP
o) MRT (computer)
p) Darts (second 12 throws)
q) Questionnaire (part B)
Descriptive Statistics for the Variables Collected in this Study
The following tables list the basic descriptive statistics for the variables discussed in this report. These values are based on data from 72 individuals, except for MENARCHE (N=71) and ARM-CIRC (N=69).
Mean SD minimum maximum
Age at testing 23.2 5.1 18.0 43.1
Time between testing
and MRI (months) 2.0 3.4 -0.1 15.8
Age difference between
sisters (years) 2.7 1.2 5.6 1.0
Table 6.2: Schooling and socio-economic status variables
variable name Mean SD minimum maximum
Number of siblings 3.1 1.2 0.0 6.0
Job SES rating: 1
subject 9.1 15.7 0.0 67.3
mother 32.5 24.9 0.0 88.3
father 47.1 27.8 0.0 88.3
family 53.9 24.1 0.0 88.3
Years of schooling:
subject SCHOOL-S 14.9 2.1 11.0 21.0
mother SCHOOL-M 14.3 4.0 3.0 30.0
father SCHOOL-F 15.4 4.8 0.0 26.0
1Based on rankings given in Stevens and Cho (1987), whose rankings range from 0 to 100. See chapter 8 for more detail.
Table 6.3: Neuroanatomical variables
Mean SD minimum maximum
BRAINVOL 1175.4 76.9 986.6 1392.7
PROSVOL 1020.0 69.2 853.9 1202.9
GREYVOL 704.1 53.0 604.7 836.6
WHITEVOL 337.4 42.3 169.7 434.8
MMMVOL 155.3 13.6 94.8 189.9
PROSVOL tissues:
% grey 69% 3% 63% 84%
% white 33% 3% 17% 39%
mean grey pixel value 62.9 6.0 48.6 77.9
SD grey pixel value 8.6 1.3 6.2 13.8
mean white pixel value 81.8 7.7 62.2 107.5
SD white pixel value 4.6 0.4 3.7 5.9
EFFECT SIZE 2.8 .27 1.8 3.3
Test variable name mean SD minimum maximum
Vocabulary VOCAB 32.5 7.6 15.0 46.0
Raven's progressive
matrices
correct RAVEN 29.5 4.3 18.0 39.0
attempted 40.3 5.6 28.0 60.0
Mental rotation
correct MRT-N 32.0 4.1 17.0 39.0
decision time (ms) MRT-SPEED 5629 1821 2417 10536
SD of decision time (ms)1 3633 1825 1062 9912
Stroop
words 110.5 13.0 76.0 138.0
colors 81.6 9.9 52.0 108.0
conflicting 51.8 8.8 30.0 76.0
adjusted for word and STROOP 5.0 7.4 -12.3 25.8
color speed
Trails
Part A (seconds) 20.3 4.5 12.9 36.2
Part B (seconds) 43.1 12.0 19.9 78.6
B minus A (seconds) TRAILS 22.8 9.8 3.7 49.6
Verbal fluency VERBALFL 45.9 10.5 26.0 67.0
Wisconsin Card Sort Test
perseverative errors WCST-PERS 14.6 12.4 3.0 52.0
Simple reaction time
decision time (ms) SIMPLE-RT 290 47 214 424
SD of decision time (ms)1 75 60 26 419
Object identification
decision time (ms) OBJECT-ID 682 157 364 1026
SD of decision time (ms)1 184 146 43 823
Sentence verification
right-branching (RB) clauses:
decision time (ms) 3522 1082 2030 8652
SD of decision time (ms)1 1064 810 314 4852
center-embedded (CE) clauses:
decision time (ms) 4214 1257 2358 8152
SD of decision time (ms)1 1218 790 251 4425
average of median RB and CE SENTENCE 3868 1125 2194 8226
verification decision times (ms) -VERIF
difference (center-embedded SYNTAX 692 661 -852 2953
minus right-branching
decision times) (ms)
1Within-subject standard deviations
Table 6.4: Cognitive tests (continued)
Movement times for
reaction time tasks Mean SD minimum maximum
Simple reaction time
mean (ms) 260 51 138 359
SD (ms) 40 10 24 83
Object identification
mean (ms) 280 73 139 576
SD (ms) 83 41 34 354
Sentence verification
right-branching clauses:
mean (ms) 257 65 118 470
SD (ms) 57 34 15 228
center-embedded clauses:
mean (ms) 256 63 118 439
SD (ms) 56 28 14 142
Mean SD minimum maximum
Oldfield Inventory
Left hand 2.0 3.6 0.0 16.0
Right hand 14.3 4.5 0.0 20.0
Handedness task
Left hand 57.5 13.7 33.0 105.0
Right hand 79.0 14.3 47.0 115.0
Total 136.5 22.0 84.0 190.0
Handedness score .16 .13 -.38 .36
Table 6.6: Sociality indicators
variable name Mean SD minimum maximum
Number of people ADDRESS BOOK 62.7 48.7 0.0 221.0
in address book TOTAL
Number of relatives ADDRESS BOOK 7.6 9.8 0.0 56.0
in address book RELATIVES
Reported number of people PEOPLE SEEN 15.5 13.6 1.0 100.0
seen socially in the SOCIALLY
average month
Reported number of people PEOPLE TALKED 35.8 46.1 4.0 227.0
talked to for more than TO
5 minutes in one week
Sociality self-rating SELF RATING 3.0 1.2 0.0 5.0
Subject's rating of sister RATING BY 3.0 1.5 0.0 5.0
SIBLING
variable name Mean SD minimum maximum
Recalled age at first MENARCHE 13.0 1.4 10.8 17.8
menstruation
variable name Mean SD minimum maximum
Average distance from target THROW 12.4 3.9 4.4 22.6
(cm)
Mean SD minimum maximum
Hours per week 4.1 3.7 0.0 18.0
Table 6.10: Anthropometric data
variable name Mean SD minimum maximum
Triceps (mm) TRICEPS 17.2 5.5 8.5 33.2
% Fat % FAT 26.2 5.5 16.1 38.2
Body Weight (kg) BODY-WT 62.2 11.2 44.0 101.2
Lean Body Weight (kg) LBW 45.6 6.5 32.3 70.7
Arm Circumference (cm) ARM-CIRC 27.6 2.8 22.2 36.0
Body Height (cm) BODY-HT 163.8 6.3 149.8 177.7
Sitting Height (cm) SIT-HT 87.6 3.0 80.9 94.8
Leg Length (cm) LEG-HT 76.3 4.5 66.8 86.4
6.2.5.2 Methods of Data Analysis
Transformation of Variables
The following scores were transformed (multiplied by -1) such that better performance was reflected by a higher score: MRT-SPEED, TRAILS, WCST-PERS, OBJECT-ID, SENTENCE-VERIF, SYNTAX, SIMPLE-RT.
This was done to avoid confusion when looking at the resulting correlations between variables, and to aid in factor analysis.
Calculation of Family Averages and Within-Family Differences
Family means and within-family differences were calculated. The family means data were of course the average values on each variable for each sibling pair. Within-family differences were calculated by subtracting one sister's score from the other, with the order of subtraction of course being consistent within sibling pairs across all variables. Because the order of subtraction within each sibling pair is arbitrary, and because correlations based on any one order of subtraction for a set of sibling pairs can vary slightly from correlations based on within-family differences calculated with a different (but equally valid) order of subtraction, a double entry method was used. For each sibling pair, (sib A) minus (sib B) constituted one set of data points, and (sib B) minus (sib A) constituted another set. Thus, for 36 pairs, there were 72 cases upon which the correlations were calculated, representing every possible way to order the subtraction of each sib pair. This method results in correlations which closely approximate the average correlation which would be obtained for all possible pair orderings (David Rowe, personal communication).
Because MENARCHE and ARM-CIRC data could not be collected from all 72 subjects, within- and between-family values were calculated only for those family pairs in which data was available for both sisters. Thus, within- and between-family correlations and statistics involving MENARCHE were based on only 35 pairs, and ARM-CIRC values were based on only 34 pairs.
Appendix C includes the intraclass correlations for all the variables used in this study. A number of these are fairly low, indicating that there is proportionately more variability within-families, and therefore that there is restriction of range in this sample (the average intraclass correlation among siblings for IQ is rÅ0.49 in large representative samples; Paul 1980).
Age Correction
All test scores and anatomical variables were corrected for age (in days) before further analyses of the data. This was done by using the residuals of a multiple regression of each variable on age, age2, and age3. Using higher powers of age ensures that any simple non-linear effects are removed (Jensen 1980). Note that this method removes only that variance which is correlated with various powers of age, leaving all other variance untouched. This process was done separately for the family means data and within-family differences data: Mean family age was used to correct the family means data, and within-family difference in age was used to correct the within-family differences data.[1] This was done because it was found that some residual age correlation remained within families even after age correcting the data set as a whole (i.e., ignoring family units and pooling all subjects together). Not all subjects had MRI's taken the same amount of time after their cognitive testing. While this difference was unlikely to have had a major effect, age-at-MRI was used to age-correct the variables collected at the MRI session, and age-at-cognitive-testing was used for the variables collected at the cognitive testing session.
Correction for Simple Reaction Time
A number of the tests given to these subjects had some form of time limit associated with them. Correlations between various elementary cognitive tasks and more complex psychometric tests of mental ability can be attenuated by individual differences in simple reaction time (SRT), presumably because SRT contains a much larger portion of variance associated with purely peripheral (i.e., sensorimotor) processes (Jensen and Reed 1990). In order to eliminate any variance associated with individual differences in SRT, all variables were also corrected for simple reaction time by calculating residuals for regression of each variable on SRT (following Jensen and Reed 1990). Again note that this method removes only that variance which is correlated with SRT. This correction was not applied to the sociality, menarche, throwing variables, which were corrected for age only.
Principal Component Analysis
As noted in the previous chapter, cognitive test scores on diverse batteries of tests are generally positively correlated, with the number and size of the positive correlations among such a battery of tests far exceeding that expected by chance (Jensen 1987). Because of this, it is not legitimate to treat each cognitive test as an independent variable, as if it they were all measuring different cognitive dimensions. It is therefore important to estimate g in any set of cognitive data for which one wishes to assess the relationships between cognitive ability and other variables. The g factor can be estimated in a number of ways, the simplest being the first unrotated principal component in a principal component analysis.
In the present analysis, the first unrotated principal component (1st PC) scores were calculated on the age-corrected, SRT-corrected variables. This was done separately for both the within-family and between-family datasets. These 1st PC scores where then regressed out of the cognitive variables and the anatomical variables, resulting in partial correlations of the cognitive variables with the anatomical variables with 1st PC scores controlled. This will allow us to assess the extent to which the cognitive variables correlate with neuroanatomical variables independent of the effect of general cognitive component.
Relationships between MRI and Cognitive Test , Sociality, Menarche, Throwing, and Anthropometric Data
Correlational Analysis
Individual within- and between- family correlations were calculated for each of the possible brain-anatomy vs. cognitive-component relationships. Between-family correlations were calculated using the means of the siblings within each family for the variables of interest, while within-family correlations were calculated using the differences between pairs of siblings for these variables (Jensen 1980; double entered as discussed above).
Significance levels were not adjusted for multiple comparisons, and therefore must not be taken at face value. Such a correction would have an effect only on the largest sets of comparisons (e.g., among all the cognitive variables, in which other statistics are presented to indicate overall significance Ð see next chapter). For the comparisons between cognitive and neuroanatomical variables, a correction can be misleading because several of the neuroanatomical variables are highly correlated, often because one is simply a subset of another (e.g., BRAINVOL and PROSVOL). One could, for example, take a significant correlation between two variables and, by taking a sufficiently large number of appropriate subsets of these variables, create a matrix in which none of the entries are technically significant after Bonferroni correction.[2]
Paired Mean Differences Analysis
A simpler (but less powerful) way to address the question of whether a relationship exists between brain volume and behavioral ability within families is to compare mean differences on behavioral dimensions for larger versus smaller brained siblings. Average differences between sibling pairs were calculated on the behavioral variables, such that the smaller brained sibling's behavioral variable score was subtracted from the larger brained sibling's score on the same variable. To the extent that brain volume is related to behavior, the mean difference between larger- and smaller-brained siblings will be positive. The significance of each of these mean differences was calculated by means of paired t-tests.
To compare age-corrected values, it was necessary to do this correction on the overall data set first (treating all 72 subjects as individuals), rather than after calculating within- and between-family values as was done for the correlational analyses (see justification for this procedure above). In addition, it was necessary to calculate first principal component scores first as well.
As a further test, the sibling pairs were divided into three groups of equal N: those who differed the most in BRAINVOL, those who differed the least, and those who fell between these extreme groups. Mean behavioral differences were then calculated separately on each of these three groups, with statistical significance again estimated using paired t-tests. The expectation is that if BRAINVOL contributes to some behavioral ability, the siblings who differed the most on BRAINVOL should tend to differ the most on that behavior.
Reliability of Measures
For some of the calculations reported below, estimation of the reliability of the measures will be needed. Reliability of the neuroanatomical variables was estimated to be greater than r=0.95 (see section 6.2.5.1 above). Table 6.11 lists the reliability estimates for the behavioral tests for which it is possible to do so. These were estimated by correlating split-halves (either odd/even or first/last) of the tests, followed by correcting the correlations with the Spearman-Brown formula (the correlation of split-halves is the reliability of a test with half as many items as the actual test given, hence the need for this correction). For these calculations, data from up to 90 subjects was available (only 72 had usable MRI's). As can be seen, most of the tasks devised for and used in this study had very high reliabilities.
Table 6.11: Test reliabilities
|
Variable |
Estimated reliability |
number of subjects |
type of split |
|
SENTENCE-VERIF: |
|
|
|
|
right-branching |
.93 |
90 |
1st/last |
|
center-embedded |
.92 |
90 |
1st/last |
|
SYNTAX |
.53 |
90 |
1st/last |
|
OBJECT-ID |
.99 |
90 |
odd/even |
|
SRT |
.94 |
90 |
odd/even |
|
MRT-SPEED |
.88 |
90 |
pseudo odd/even |
|
MRT-N |
.70 |
90 |
pseudo odd/even |
|
|
|
|
|
|
THROW |
.67 |
88 |
1st/last |