Auditory Verbal Learning Test
Abbreviation: AVLT
Background
Learning and later testing of word lists was certainly known as a method for assessing memory performance since the second half of the 19th century, and presumably even earlier. The form of such assessments probably most frequently used today, namely having an examinee learn a list of about 15 words several times in succession and recall them from memory after a distraction, was first published in 1919 by the Geneva psychologist Claparède (see1)). His student André Rey carried out clinical investigations with this memory test in the 1940s2). A detailed description of the (modified) procedure appeared only much later in a book publication3). At that time, the test was being used in translation by clinicians in other countries without much being published about it.4)
The procedure was introduced into American psychology under the name Auditory Verbal Learning Test (AVLT) initially by Taylor5). Through its inclusion in Lezak's handbook of neuropsychological assessment6), it later became generally known. Today's versions of the AVLT (including the published German version) rely largely on the procedure used by Taylor.
Today the AVLT is one of the most frequently used memory tests worldwide. It is discussed in all textbooks and handbooks of neuropsychological assessment7) 8) 9). In 1996, Schmidt compiled the American empirical literature on the AVLT in a monograph.10)
German versions of the AVLT were first presented in journal articles as the VLMT11) and the AVLT12), respectively. The VLMT appeared in 2001 as a complete test procedure in German language ready for use13).
In what follows, we rely predominantly on the VLMT. It is probably also the most widely used German form of the AVLT. For the norms, however, we will draw on further literature.
Test material
If one wanted to administer any “auditory word learning test”, one would need only 45 nouns (classically ones that describe real objects). The first 15 form a list that is learned over five trials, the next 15 are used as a so-called interference list and are learned once afterwards. Then the words learned first must be recalled. Finally, the last 15 of the 45 words together with the first 15 form a list of 30 words in which the words learned first must be recognized.
Given the advantages offered by the standardization of the test material, it is in most cases highly advisable to fall back on published versions where such exist. Since 2001, a published German-language version has been available in the form of the Verbaler Lern- und Merkfähigkeitstest (VLMT), which is easy to administer and fulfills all criteria of objectivity of the administration conditions. In addition, besides the classical test scores, the VLMT allows a number of further supplementary scores to be obtained, which, however, are not presented further in TDB2Online. Even without this added benefit, we think that the objectivity of the administration conditions and the convenience of the ready-to-use presentation are in any case worth the price of the professional test material. For the VLMT, it consists of the test manual, record sheets, and result sheets.
Test administration
Over the many decades that the AVLT has existed, various ways of administering the test have naturally become widespread. However, one of them has established itself as a quasi-standard, and the VLMT follows it as well:
There are two word lists of 15 nouns each for presentation, one of which (A) is the learning list, the other (B) the interference list. The recognition list consists of the two lists A and B and a further 20 distractor words, of which 10 are semantically and 10 phonetically similar to the words in A.
List A is read aloud five times. After each reading, the examinee is to recall as many words as possible from memory. This part of the test yields (at least) two raw scores for learning performance, namely the number of words reproduced from memory across all five trials (Dg1-5 in VLMT terminology) and the number of words reproduced in the last trial (Dg5). Then the interference list is read aloud once; as before, it is to be reproduced immediately afterwards. Subsequently, the examinee is to recall once more - without renewed reading - the words of list A learned first. The number of words recalled here yields a first raw score for memory performance; in the VLMT this measure is simply called Dg6, and in many English-language publications it is also cited as Post-Interference Recall (PIR). After 30 minutes, which are to be filled with nonverbal tests, a free recall of list A again takes place without renewed presentation. The number of words recalled here is the raw score for delayed recall (Dg7 in the VLMT, otherwise also often Delayed Recall). Directly afterwards, the recognition list is read aloud and the examinee must answer yes if he or she recognizes a word from the originally learned list A. In most procedures, the corresponding measure is the number of correctly recognized words (Recognition Memory or Delayed Recognition). This measure also exists in the VLMT, but there the number of correctly recognized words minus the number of errors is the preferred measure.
Test raw scores
The AVLT, and thus the VLMT as well, traditionally yields the number of correctly recalled words as the raw score for each recall trial.
Not all of these raw scores are equally important. The five learning trials, for example, are mostly considered only in their sum, that is: how many words an examinee was able to recall across all five learning trials. A useful raw score is also the number of words recalled in the last learning trial, because it gives the best estimate for the post-interference trial.
The following measures are in international use:
For learning performance (both measures correlate very highly with each other):
- Number correct across all five learning trials (in the VLMT: ΣDg1-5)
- Number correct in trial five (Dg5)
For immediate recall:
- Number correct in spontaneous recall directly after the recall of the interference list (Dg6)
For delayed recall:
- Number correct in recall after 30 minutes (Dg7)
For delayed recognition:
- Number correct in recognition after 30 minutes (W)
- occasionally also: number correct minus errors (W-F)
All of these measures are recorded in the VLMT. However, not all of them count among the favored measures there. On the basis of factor-analytic results, the VLMT prefers relatively “pure” measures for the factors learning, consolidation, and recognition. These are the total learning performance ΣDg1-5 for learning, the loss after interference (Dg6-Dg5) or after a time delay (Dg7-Dg5) for consolidation, and the corrected recognition performance (W-F) for recognition. Such difference or percentage scores do occur in the international literature, but the most varied variants are found there, which makes meta-analytic processing impossible. Only the parameters Dg5, Dg6, Dg7, and W show a high overlap, which is why we concentrate on these four in TDB2Online.
Distributional properties of the raw scores
To our knowledge, there are only two publications in the literature that report more extensive, age-specific percentile distributions of test results on the AVLT. The first source is the German VLMT manual14), which extends similar earlier data from the working group15), and the second is a publication by Ivnik et al. (1992)16), which reports data for older patients. From the percentile data in these publications, raw score distributions could be back-calculated for the four measures considered in TDB2Online. They are shown in Figures 1 to 4.
Figure 1: Back-calculated raw score distribution for learning performance in the fifth trial, see text)
For the parameter Dg5 (Figure 1), only data from the VLMT manual were available. Unfortunately, Ivnik et al. (1992) provide neither data for Dg5 nor for ΣDg1-5. Instead, the authors provide data for “Learning over Trials”, a parameter that corrects overall learning performance (ΣDg1-5) for performance in the first trial. Sometimes one wishes that test authors were less creative in devising new parameters.
For the distribution of learning performance in the fifth trial, a massive ceiling effect can be noted, at all three age levels. If young adults on average can freely recall more than 13 of 15 words after five learning trials, this has to be the case. The ceiling effect could be remedied only by considerably longer word lists, which, however, would certainly lead to longer testing times and presumably also to frustration in lower-performing examinees. Since such word lists are generally used more for measuring deficits than for measuring potential, one can live with the ceiling effect given the shortened testing time. Similar tests used in the dementia field manage with even 10 words (CERAD, RBANS).
The low test ceiling has consequences for the linearity of the measurement scale, which no longer holds at the upper end. Unfortunately, there is no reliable method to correct this, because only a very limited amount of data from relatively young persons is available to us. For other test procedures (such as the HCT or TMT), there were larger samples and a larger age range, both of which affect the quality of the linearization.
Figure 2: Back-calculated raw score distribution for recall performance in recall after interference, see text)
Distribution data on the parameters of recall performance (Dg6 and Dg7, Figures 2 and 3) were available from both studies, which yields an age range extending into the 80s. For both measures, the distributions are generally quite symmetrical. Only in young adults (age group 15-29; the VLMT does not break this down more finely) is a certain ceiling effect found here as well. A linearization would be possible here, but it would represent a great effort for a limited yield.
Figure 3: Back-calculated raw score distribution for recall performance in recall after 30 minutes, see text)
Figure 4: Back-calculated raw score distribution for recognition performance, see text)
The situation is different for recognition (Figure 4). Here the ceiling effect is enormous. Here one would also have the data of the older examinees, which are needed in order to map low and very low performance of young examinees correctly in performance scores. Here too, however, the question arises of the ratio of effort to yield. Measures in recognition can usually be interpreted meaningfully only when larger differences between recall and recognition appear. This can presumably also be established with the non-linearized scaling used here, even though it does not adequately reflect the underlying performance.
Linearization of the measurement scale
For the reasons mentioned in the last section, further linearization of the measurement scales for Dg5 and W, which are certainly nonlinear at the upper end, was dispensed with. If more extensive norming data for the VLMT become available later, the procedure could be reconsidered and changed if necessary.
Norming studies
For the creation of the TDB2Online performance norms, recourse was systematically taken to as many accessible studies as possible that contained AVLT data of healthy examinees. Even though the AVLT is a test based on verbal material, a comparison across language boundaries seemed possible to us. All currently existing versions of the AVLT use concrete words of the respective language. With few exceptions, they are even the same objects in the different languages.
The condition for the inclusion of a study was that means and standard deviations were reported for as many as possible of the four raw scores included in TDB2Online, and moreover for a sample of healthy examinees in several differentiated age ranges. In the well-known collection of normative data for standard neuropsychological procedures, published as a monograph in a second edition in 2005 under the title Handbook of Normative Data for Neuropsychological Assessment (Mitrushina et al., 200517) ), 24 studies are listed for the AVLT, of which 9 could be used for the TDB2Online norms. These are studies 218), 419), 720), 921), 1122), 1223), 1524), 1725), and 2026) in the order of listing in the handbook. In principle, two further studies would also have met the inclusion criteria, 1927) and 2428). However, both overlap with other studies already included (19 with 12 and 24 with 11), and although in both cases the included studies represent only a subsample of the non-included ones, they, unlike the non-included ones, each cover all four raw scores.
Including the VLMT manual, 10 studies could thus be located in which itemized normative data were given for age ranges between 15 and 97 years. The number of included examinees ranges, depending on the raw score, between 1797 (trial 7) and 2839 (trial 5) persons. Table 1 shows the most important characteristics of the included studies. By means of a meta-analytic summary, these norms were integrated and thereby made independent of the peculiarities of the individual study.
Table 1: Overview of the meta-analytically pooled norming studies
| First author | Year | Age range | N | Education | IQ | % male | Country |
|---|---|---|---|---|---|---|---|
| Query | 1983 | 19-81 | 677 | 11 | 94 | 100 | USA |
| Cohen | n.a. | 60-89 | 85 | 14 | n.a. | 35 | USA |
| Wiens | 1988 | 20-49 | 209 | >12 | 111 | 87 | USA |
| Nielson | 1989 | 20-54 | 101 | n.a. | 99 | 53 | Denmark |
| Geffen | 1990 | 16-86 | 153 | 11 | n.a. | 50 | Australia |
| Ivnik | 1990 | 55-97 | 390 | 13 | 107 | 37 | USA |
| Mitrushina | 1991 | 57-85 | 156 | 14 | 117 | 40 | USA |
| Selnes | 1991 | 25-54 | 696 | 16 | n.a. | 100 | USA |
| Savage | 1992 | 20-76 | 134 | >11 | n.a. | 50 | USA |
| Helmstaedter | 2001 | 15-79 | 276 | n.a. | 108 | 68 | Germany |
Meta-analytic summary of the norms
Figure 5 shows the age courses of the raw scores in trial 5, that is, the last learning trial, for the 10 studies that provided data for this. In total, the data of 2,839 persons enter into the figure.
Figure 5: Age course of the raw scores in trial 5
Compared with other procedures and other raw scores, the results of the 10 studies are relatively similar. At the lower performance margin lie the data of Query and Cohen, at the upper margin those of Helmstaedter for the VLMT as well as those of Wiens and Selnes. The curve labeled Mquer shows the averaged raw scores across the age course. For this averaging, the studies were weighted according to their sample size. If the whole is additionally smoothed with a moving average across the age groups, the result is the curve labeled Mglatt (brown line with circles). The two other thicker brown curves show the plus/minus 1-sigma limits, where the standard deviations of the individual studies used for this were likewise averaged across the studies and smoothed. In the young, higher-performing age groups, the standard deviations were systematically lower. This is explained by the restriction of the possibility of variation due to the ceiling effect, which carries more weight the closer the average performance approaches the upper margin.
The age-associated decline in performance flattens out again in these data from an age of 85. This is implausible, and performance of those over 84 is presumably overestimated here. In this range, the mean is based on only very small sample sizes and, from 90 on, on only a single study. In this age range, the AVLT is probably also rarely used. In general, the choice here is likely to fall on word lists with only 10 elements, as in RBANS or CERAD, for example.
Figure 6 shows the age courses of the raw scores in trial 6 (this is the trial after the interference trial) in the 9 studies that provided data for this. These raw scores measure spontaneous recall performance after the brief interference by the distractor list. The 9 studies summarized the results of a total of 2,462 persons.
Figure 6: Age course of the raw scores in trial 6
Compared with the last learning trial (Figure 5), the absolute recall performance changes, but not the relative position of the individual studies. The same studies as in Figure 6 are also distinguished here by above-average or below-average performance. The weighted means summarize the study results well; there are no gross outliers.
The smaller standard deviation in the young cohorts caused by the ceiling effect in trial 5 is (almost) not seen in trial 6. Here, performance, at around 11 words, is far enough from the maximum of 15 that there are hardly any ceiling effects.
Figure 7 shows the age courses of the raw scores in trial 7, the spontaneous delayed recall of the previously learned list after about half an hour. Only 7 studies provided data for this. They are based on the results of 1,797 persons.
Figure 7: Age course of the raw scores in trial 7
If one compares the course of performance in delayed recall (Figure 7) with that in recall after interference (Figure 6), one finds that the course in Figure 7 is initially rather flat and declines more steeply only from an age of 60 years. One should not overinterpret this. In Figure 7, the data from the study by Query & Megran29)(1983) are missing, which showed the clearest age-related decline at relatively early ages in both learning performance and immediate recall. Here the shape of the course probably rather reflects the different composition of the study samples.
Figure 8 shows the age courses of the raw scores in delayed recognition after about half an hour. Here again there were data from 9 studies, based on the results of 2,738 persons.
Figure 8: Age course of the raw scores in delayed recognition
As the distribution data in Figure 4 already showed, good recognition performance occurs in all age groups. For the upper performance range, recognition performance differentiates far too little. In the lower range, by contrast, it is quite informative. For the very slight age-associated decline in performance from about age 65, I have no explanation. In any case, the finding is not based on the results of a single study alone.
Overview of performance norms and age norms
The figures in this section summarize the performance norms and age norms of the four learning and memory test scores as they appear in TDB2Online. Peculiarities are commented on in each case.
Figure 9 shows, in a composite graphic, the relationship between raw scores and performance scores as well as the (average) influence of age on learning performance in trial 5 of the AVLT. In this and all following figures, the vertical marks indicate the performance scores from 40 to 145. The raw scores are plotted in their correct positions on the black horizontal line. The alternately red and blue lines contain the norm limits for all age groups in the overview. The five markers on each line represent the percentile ranks 2.5, 16, 50, 84, and 97.5. One such line in each case, namely the one corresponding to the examinee's age group, is drawn (in a different form) in the TDB2Online profile sheet to aid in the individual interpretation of the test results. Because the performance scores are calculated from the results of the 20-24-year-old young adults, 4 of the 5 markers of the 20-24 age group enclose the performance scores 70, 85, 100, 115. Because the test ceiling in trial 5 is far too low and ends at a raw score of 15, this does not hold for the 5th marker. Owing to the raw score, it lies at a performance score of 119, because higher performance scores cannot exist.
As explained in the section on linearization of the measurement scale, linearization was dispensed with for all four parameters of the AVLT. It would not have been able to remedy the ceiling effect in trial 5 either.
The age-associated decline in performance from young adulthood to old age amounts to roughly two standard deviations. That the only slight decline beyond 80 is probably an artifact was already mentioned above.
Figure 9: Performance scores in trial 5 with raw scores and age norm limits (see text)
The two parameters for spontaneous recall, trial 6 immediately after the interference list (Figure 10) and trial 7 after half an hour (Figure 11), can be considered together; they hardly differ with regard to their norms. Here the age-associated decline in performance amounts to somewhat more than two standard deviations. That the shape of the decline differs in the two measures is probably attributable to the fact that one large study (Query & Megran,30) 1983) provided values only for trial 6, but not for trial 7.
Figure 10: Performance scores in trial 6 with raw scores and age norm limits (explanation of the figure in the text for Figure 9)
With the parameters Dg 6 and Dg 7 as well, very good performance cannot be measured, at least not in younger examinees. The measurement scale ends at a performance score of 125 (Dg6) and 123 (Dg 7), respectively.
Figure 11: Performance scores in trial 7 with raw scores and age norm limits (explanation of the figure in the text for Figure 9)
For delayed recognition, the measurement scale already ends at a performance score of 113. Better performance cannot be measured; the measurement scale is very coarse overall, as can be seen from the division of the raw scores. One raw score point more or less in each case means a difference of 11 IQ-normed performance score points, which is more than 2/3 of a standard deviation. (For Dg 6 and Dg 7, the basal measurement precision determined by the raw score is somewhat better; there, one raw score step corresponds to roughly 6 performance score points).
Figure 12: Performance scores in delayed recognition with raw scores and age norm limits (explanation of the figure in the text for Figure 9)
Discussion
The AVLT and VLMT exist because in many clinical situations they allow a good assessment of the memory performance of adults. They are pragmatic instruments with high content validity. Patient and recipient of the report understand at once what is being measured. The results can often be communicated already as raw scores. Once someone has understood the procedure and has discussed a few test results with a neuropsychologist, he or she can usually quickly form an opinion on the meaning of a raw score result.
Psychometric purists, by contrast, have problems with this procedure. The measurement scale is far too coarse already at the level of the raw scores. The test ceiling is so low that very good performances can never be measured, and good performances can hardly be measured in delayed recognition. The four parameters are highly intercorrelated. It is best not even to think of the condition of serial independence of test items.
We have nevertheless included the AVLT in TDB2Online. The aim, as always, was the optimal visualization of a test result. For the AVLT, too, the transformation of raw scores into performance scores works well. It has two advantages: One sees at once how finely the test can resolve and how the results would have changed had the examinee recalled one more word. Through the manner of presentation, one is immediately aware of the test ceiling. In addition to these general characteristics, the age-normed scores provide the necessary orientation for clinical interpretation.