Language is vitally important to all aspects of a child’s development.  Research has shown that it is related to social (Craig, 1993; Gallagher, 1993; Rice, 1993), cognitive (Brown, 1973; Chapman, 1992; Dromi, 1993; Flavell, 1985; Schwartz & Murphy, 1975; Silva, 1980), and educational (Silva, McGee, & Williams, 1987)  domains of development.  In light of the importance of language to a child’s overall development, the early detection of disorders in any domain of language becomes crucial, as does early intervention (Silva et al., 1987).

1.1 Language Impairment And Development

        Language is central to all communication and provides the foundation for learning and socialisation.  With respect to social development, children need to be competent language users to make and maintain friendships, and to communicate effectively with peers, parents, and teachers.  Language is the primary means of establishing and maintaining social relationships (Prutting, 1982; Schwartz & Murphy, 1975). Social interaction is then a vehicle for learning and refining language skills (Craig, 1993; Gallagher, 1993; Schwartz & Murphy, 1975).  The competent language user has the potential to continually improve his or her communication ability, as establishing social relationships is facilitated by strong language skills, and the social relationships themselves provide opportunities for further language learning.  Language-impaired children; however,  may have difficulties interacting with their peers, and consequently are at risk for social rejection (Craig, 1993; Gallagher, 1993; Rice, 1993).  This rejection may then reduce their exposure to language and therefore limit their opportunities to practice and improve their conversational skills (Rice, 1993).
        A similar, reciprocal relationship exists between language development and cognitive development.  The organisation of a child's experiences and the acquisition of cognitive abilities are associated with the development of language (Flavell, 1985; Kamhi & Catts, 1989; Schwartz & Murphy, 1975).  In a review of studies investigating cognitive attainments in language impaired children, Rescorla and Goosens (1992) concluded that children with impaired language perform more poorly than peers with normal language skills on a wide range of cognitive tasks.  In addition, once a child enters school, language is the means through which all subjects are taught (Berlin, Blank, & Rose, 1980; Schwartz & Murphy, 1975).  Even subjects not obviously related to language, such as math and science, are taught through the teacher’s use of language to convey information to students.  Thus, a child's language difficulties have the potential to interfere with academic success in all subject areas.  Indeed, Brown and Edwards (1989) concluded that early language impairments later resurface as difficulties in reading and writing.  Research has shown that a language delay identified in the preschool years has a strong predictive relationship to later learning difficulties (Silva, Williams, & McGee).  Research also has shown that preschool language ability was the best single predictor of later reading ability (Aram, Ekelman, and Nation, 1984).  Maxwell and Wallach (1984) reported that a persistent language delay in the preschool years was associated with risk for chronic deficits in language and academic achievement.
        In addition to language delays being predictive of later academic difficulties, research suggests that the effects of a language delay are broad and long-term.  Weiner (1985) reviewed 17 longitudinal studies of children with speech and language impairments.  Weiner’s  review suggested that an impaired language system will have life-long negative consequences on social, academic, and vocational success.  Problems in adolescence were identified by  Donahue and Bryan (1984) who found that adolescents who had been identified with communication impairments as children were less popular with peers.  Aram et al. (1984) also reported that adolescents who had been identified with communication impairments were described by their parents as having more behaviour problems than their peers.  In addition, teachers were more likely to express concerns about their level of academic achievement.  Lunday (1996) reported findings of a 1991 U.S. Department of Labor study that expressed similar concerns, concluding that communication disorders are likely to impair the ability of adolescents and young adults to acquire job skills, find employment and remain employed.

1.2 Early Intervention

        In order to prevent or reduce many of the long-term consequences of language impairments, intervening with language impaired children as early as possible seems to be one solution.  It has been suggested that the longer speech and language delays are left without intervention, the harder it will be to rehabilitate the language system (Eisenson & Ingram, 1972; Menyuk 1975; Semel & Wiig, 1975).  Researchers have theorised that an impaired language system becomes increasingly inefficient and rigid with age and becomes less amenable to remediation.  Early intervention may serve to  modify and repair the language system at an age when it is less rigid.  This is not to suggest that later intervention is not worthwhile and should not be undertaken.  It should be noted; however, that as age increases, the difference in abilities between a child with impaired language and his or her peers widens (Wiig & Semel, 1984), possibly due to the changing language demands of the curriculum (Paul, 1991; Schwartz & Murphy, 1975).  Thus, as a child gets older, he or she must work harder to catch-up to peers.  Research has shown that the older a child is when a language impairment is identified, the poorer the outcome is for the child (Aram et al., 1984; Schery, 1985; Schwartz & Murphy, 1975).  In summary, the age at which a child begins intervention can influence the impact of a language delay on social and cognitive skill development, educational achievement, and employment potential.
        The first step in intervention with language impaired children is adequate and accurate assessment (Keith, 1994).  According to Cox and Cox (1981), accurate assessment of developing language skills in preschool children should be a mandatory skill for all practising Speech-Language Pathologists.  Clinicians make use of both standardised language tests and language sampling techniques, as well as informal, clinician-devised methods and their clinical judgement when assessing the language abilities of children (Allen, Bliss, & Timmons, 1981; Wilson, Blackmon, Hall, & Elcholz, 1991).  The use of norm-referenced, standardised tests is generally considered appropriate and necessary for identifying speech or language impairments (Schery, 1981; Stark, Tallal, & Mellits, 1982).

1.3 Standardised Assessment

        Plante and Vance (1995) state that the primary purpose for administering a norm-referenced test is to determine if an impairment is present or not.  As well as identifying if a problem exists, determining the goals of intervention and planning procedures for intervention are also objectives of assessment (McCauley & Swisher, 1984a).  Carrow-Woolfolk and Lynch (1982) further state that language assessments are concerned with identifying if there is a problem, as stated above, determining the significance or severity of the problem if one exists, describing the nature of the problem, and describing the overall language profile of the person with the problem.  To determine the existence and severity of a language problem, normative data must be used in order to compare the individual to his or her peers.  Describing the nature of the problem requires naturalistic observation and interviews with parents or teachers (Carrow-Woolfolk, & Lynch, 1982).   Clinicians often rely on standardised tests to provide all of the necessary information about the child and the disorder.  They often use tests that are familiar to them and that are easily available.  Assessment batteries; however, should be determined by the need for specific information to be used for a definite purpose, not by familiarity, ease of testing, or availability (Carrow-Woolfolk & Lynch, 1982).  Further, solely using available standardised tests may not provide a representative view of the child’s ability to use language across different meaning contexts (Crais, 1995).
        In the U.S., the use of standardised tests by clinicians to identify individuals with language disorders and to plan treatment has been increasing since PL 94-142 was introduced in 1975 (Danwitz, 1981; Plante & Vance, 1994), requiring the use of standardised tests during language assessments (Wilson et al., 1991).  In a study by Wilson et al. (1991), the Expressive One Word Picture Vocabulary Test, Clinical Evaluation of Language Fundamentals, Structured Photographic Expressive Language Test, Test of Language Development - Primary, Illinois Test of Psycholinguistic Abilities, Preschool Language Scale, the WORD Test, and the Detroit Test of Learning Aptitude were among the top ten tests used to assess expressive language.  A large majority of clinicians in the study used formal standardised tests as part of their language assessments.  Many clinicians also reported using standardised tests in conjunction with informal assessments and language sampling methods.  A primary constraint on the clinician’s assessments was time.  Half of the respondents reported that they did not have enough time to perform adequate assessment.
        Standardised tests are frequently used by speech-language pathologists because of hospital, school, or treatment centre guidelines.  Standardised test results are often required to determine a child’s eligibility for special programs and services (Schery, 1981; Wilson et al., 1991).  Standardised tests with normative data allow the speech-language pathologist to compare a child's score to a group of other children who are believed to be similar to the child on important variables, such as age, ethnicity, socio-economic status (Kelly & Rice, 1986).
        Two types of tests are used to assess children: those designed to screen for potential speech and language impairments, and those designed to identify impairments (Plante & Vance, 1995).  More than 150 tests have been developed to evaluate  various aspects of a child's linguistic performance (Wilson et al., 1991), including over 50 standardised screening instruments (Sturner, Layton, Heller, Funk, & Machon, 1994).  Standardised tests do have advantages over non-standardised assessment techniques.  Standardised tests allow for increased objectivity by the examiner, the opportunity to repeat the assessment with the same child or other children, and they reduce unnecessary and uncontrolled variation during assessment (Kelly & Rice, 1986).  However, the use of standardised tests is subject to limitations.  The structure of standardised tests and the contrived nature of their tasks limits their ability to capture how a child uses his or her language ability in everyday situations.  Many of the available tools for assessment require at least some degree of subjectivity on the part of the examiner (Cox & Cox, 1981).  Finally, despite the large number of available tests, few meet the psychometric criteria for standardised tests (McCauley & Swisher, 1984a; Plante & Vance, 1994).
 
        1.3.1 Psychometric Characteristics of Tests

        In a review of 30 standardised preschool language tests, McCauley and Swisher (1984a), found only one test that met greater than six of the 10 psychometric criteria evaluated.  Similarly, of the 21 standardised preschool language tests evaluated by Plante and Vance (1994), only four tests met six or more of the same 10 psychometric criteria.  Only one of these four tests was found to provide "acceptable accuracy in discriminating between the children with normal and the children with impaired language" (Plante & Vance, 1994, p. 15).  In both studies, the most frequently unmet psychometric criteria were the inclusion of empirical evidence of validity and reliability.  Three kinds of validity have been acknowledged as important for tests that measure any kind of behaviour: construct, criterion-related, and content validity (McCauley & Swisher, 1984a).
        A construct is “an informed, scientific idea ‘constructed’ to describe or explain behaviour” (Cohen, Swerdlik, & Smith, 1992, p. 175).  Construct validity seeks to judge the appropriateness of inferences made about an individual’s ability on a certain construct, based on test scores.  It is examined by looking closely at test authors’ description of the construct they claim their test measures and comparing it to the actual test content (McCauley & Swisher, 1984a).  Criterion-related validity looks at correlations between scores on the test under scrutiny and scores on another test measuring the same behaviour (McCauley & Swisher, 1984a).  Convergent correlations provide evidence of criterion-related validity, as shown when test scores correlate with scores on older, more established and validated tests measuring the same or a similar construct (McCauley & Swisher, 1984a).  Alternatively, divergent correlations provide evidence of criterion-related validity, as is shown when test scores do not correlate highly with other tests of behaviours with which they should not theoretically be correlated (Cohen et al., 1992).  Content validity refers to how well a test samples behaviours that are representative of the whole family of behaviours, or the construct, the test was designed to measure (Cohen et al., 1992).  Therefore, a general language test would be expected to sample behaviours from all of the language components.
        Factor analysis can be used to identify clusters of variables, or factors, that may be causing test scores to correlate (Cohen et al., 1992).  Exploratory factor analysis (EFA) seeks to identify the sources of influence within a correlational matrix (Bryan & Yarnold, 1995).  That is, the items in the matrix are grouped according to theoretical factors or similar underlying constructs, which must be identified by the researcher, that lead to the correlations between the items (Bryan & Yarnold, 1995).
        The most important psychometric property of a test, according to Plante and Vance (1995), is how accurately a test determines whether or not an impairment is present.  Standardised language tests should provide evidence of their ability to distinguish children with normally developing language from children with impaired language development.  Tests should also be developmentally sensitive.  That is, tests should reflect the development of language ability as children grow older.

        1.3.2 Developmental Sensitivity

        A language test’s sensitivity to a child’s development is very important.  Evidence of changes that are associated with age for certain constructs has been accepted as evidence of construct validity.  It is argued that the manifestation of the construct changes with age and the scores on the test change with age, therefore the test is measuring that construct (Cohen et al., 1992).  Although developmental sensitivity is not enough to establish validity on its own, it adds to the validity and is an important aspect of developing any test for use with children.
        For tests that claim to measure constructs that are expected to change as a child matures, such as language ability, test scores should change over time as well (Cohen, et al., 1992).  For example, if a three year old, a four year old, and a five year old all completed the Peabody Picture Vocabulary Test - Revised (Dunn & Dunn, 1981), one would expect that the raw scores received would be highest for the five year old and lowest for the three year old, because vocabulary increases with age.
        Unfortunately, many test manuals do not include information about how test scores relate to chronological age.  The Structured Photographic Expressive Language Test-Preschool (SPELT-P) manual (Werner & Kresheck, 1983) reports that “the structures measured by the SPELT-P are developmental in nature and performance should be related to chronological age” (p. 19).  Mean scores on the SPELT-P increased with age in the standardisation sample, but no correlations were reported of raw score with age were reported, and no mention was made of the significance or non-significance of the increases.
        The Renfrew Action Picture Test (Renfrew, 1993) also lists mean scores for the Information and Grammar subtests for age ranges from 3;6-3;11 to 8;0-8;5.  Mean scores increased with age, but no information as to the statistical significant of the increases was mentioned.
        The Reynell Developmental Language Scales (Reynell, 1978) is based on three components of expressive language (structure, vocabulary, and content) that are developmental in nature.  According to the manual, language structure develops rapidly over the first three years of life, and by the age of five, most of the basic language structure is available for use.  In the development of the scales, the development of the individual components was taken into account.  For example, vocabulary develops very rapidly between 1;6 and 4;6; therefore, greatest coverage was given to this age range when vocabulary items were chosen for the test.

        1.3.3 Relationships between language tests

        Other than correlational information, little data exist about the relationships between language tests (Howlin & Cross, 1994).  Much of the data come from standardisation samples, and often very small samples of younger age groups are all that are available (Howlin & Cross, 1994).
        A study by Howlin and Cross (1994) examined the scores of 35 three to four year olds on six different standardised language tests commonly used by speech-language pathologists in Great Britain.  They found correlations between age-equivalence test scores that ranged from .32 to .82.  An age-equivalence test score is the mean score obtained by a group of children at a given age.  When they correlated age-equivalent test scores with chronological age, the correlations ranged from .04 to .39.  Significant correlations among age equivalence scores and chronological age were expected, although not found.  The restricted age range in the study was determined to be the reason for the small correlations (Howlin & Cross, 1994).
        Harwood (1997) examined  correlations among eight language tests for a mixed group of children with normal language and children aged three to six with impaired language.  She found correlations among raw scores on the tests that ranged from 0.16 to 0.86, and correlations among z scores that ranged from .11 to .83.
Language test manuals often report correlations between test scores and other similar tests as evidence of concurrent validity.  For examples of these correlations as well as reliability correlations, please see the individual test sections in Chapter 2.

        1.3.4 Ability of Tests to Distinguish Children with Normal Language from Children With Language Impairment

        Important decisions regarding the provision of clinical services are dependent on the results of language assessments.  It is critical that tests meet psychometric criteria and distinguish between children with language impairment and children without language impairment children (Plante & Vance, 1994).  In order to identify a child with language impairment, language performance must be assessed in more than one context, including contexts that stress the language system (Lahey, 1990).  Tests designed to identify language disorders should discriminate between children with normal language and children with language impairment with a high degree of accuracy.  According to Plante and Vance (1994), 90% is considered to be good accuracy, while 80% accuracy is only considered to be fair.
        The accuracy of current tests that are used to classify children as  language-impaired or non-language-impaired has been called into question (Howlin & Cross, 1994; Howlin & Kendall, 1991).  More information about the psychometric properties of these tests, such as the constructs these tests are actually measuring and the items that best represent the constructs, needs to be obtained for these tests.
        According to Plante and Vance (1995), the diagnostic accuracy of individual  tests is dependent on an empirically derived cut-off score for that test.  In their study, simply adopting a standard cut-off score, such as 1, 1.5, or 2 standard deviations below the mean, on the Clinical Evaluation of Language Fundamentals-Preschool (CELF-P) and SPELT-P resulted in reduced diagnostic accuracy of the tests.  They found that the two tests were maximally effective at distinguishing between normal and abnormal language learners at different cut-off scores.  The cut-off scores were -0.27 and -1.39 standard deviations for the CELF-P and SPELT-P, respectively.
        It should be noted that speech-language pathologists (SLPs) do not simply accept the results of standardised tests, but that they bring their years of experience to the assessment situation as well and examine information other than test scores, such as observations made during assessment.  Experienced SLPs often report knowledge of individual tests and are aware of some test idiosyncrasies that they have observed through frequent use of the tests (Warr-Leeper, 1997, personal communication).  As a result, children who fail a language test are not necessarily labelled as language impaired by clinicians, and children who pass a language test may still be diagnosed with a language impairment (Allen et al., 1981).

1.4 Language Sampling

        Standardised tests access language ability through contrived tasks that may have little resemblance to real environments (Hughes, Fey & Long, 1992).  In order to gain an understanding of a child’s language ability in a more natural communicative setting, language sampling techniques have been developed (Klee, 1992).  Language sampling is appealing to clinicians because it is the most ecologically valid way of assessing a child's ability to use language communicatively (Paul, 1995).  Interacting and communicating with a child gives the clinician a clear impression of how that same child interacts with peers, parents and teachers.  In addition, the clinician is provided with spontaneous examples of a child's language ability.  Data from the language sample can then be used by clinicians to generate hypotheses about the child's language system and how it may be impaired (Lahey, 1988).  McCauley and Swisher (1984b) note that the detailed description of a child’s language production that is provided by a language sample analysis is a good source of intervention goals.  They also state that even though language sampling techniques may be lengthy and difficult, the information they provide is valuable enough to outweigh the difficulties.  Furthermore they suggest that language sampling analysis is a preferable source for determining therapy objectives compared to standardised tests.  Thus it appears that language sampling measures are considered valuable for both identification of a language impairment and selection of intervention objectives.
        A variety of language sampling methods have been developed.  They include: Assigning Structural Stage (Miller, 1981), Index of Productive Syntax (IPSYN; Scarborough, 1990), Language Assessment and Remediation Procedure (LARSP; Crystal, Fletcher, & Garman, 1976), Developmental Sentence Scoring (DSS; Lee, 1974) and Systematic Analysis of Language Transcripts (SALT; Miller & Chapman, 1993).
SALT is a computerised method of calculating mean length of utterance (MLU), total number of words (TNW), number of different words (NDW), and calculating frequency counts of specific morphemes.  Other computer programs designed to analyse language samples are also available on the market (e.g., Computerised Profiling: Long, 1987; Child Language Data Exchange System: MacWhinney, 1995).  The output from such computer programs may result in many numbers quantifying a language sample (e.g., total number of utterances, total number of complete and intelligible utterances, total number of words) but the meaning and value of these numbers require more research (Klee, 1992).
        Klee (1992) examined nine measures and found three of the language sample measures had good developmental and diagnostic characteristics.  They were Mean Sentence Length (MSL), TNW, and NDW.  MSL, TNW, and NDW correlated either moderately or highly with age for both normal and specific language impaired (SLI) groups and showed potential for assessing change in language production over time.  MLU was shown to have developmental and diagnostic characteristics in an earlier study (Klee, Schaffer, May, Membrino, & Mougey, 1989) and was correlated with age at r ‘ .75 for 24 normally developing children from 24 to 50 months of age, and at r ‘ .77 for 24 language impaired children from 24 to 50 months of age.  In other studies, MLU was found to be correlated with age with correlations ranging from .16 (Klee & Fitzgerald, 1985) to .88 (Miller & Chapman, 1981).

1.5 Purpose of the Present Study

        The purpose of the present study was to examine eight language tests commonly used by speech-language pathologists in Southern Ontario.  The following questions were addressed:

1. Are the tests developmentally sensitive?  Overall, do the test scores correlate with the chronological ages of the
    children?

2. Does developmental sensitivity differ across various age groupings?  That is, are correlations with chronological age
    different for three year olds, four year olds, and five year olds?

3. What are the relationships among the tests?  How do they correlate with one another?

4. What may account for the relationships among the tests?

5. Are the tests able to distinguish between children with normal language skills and children with impaired language skills?

6. Which tests best discriminate between normal language learning and language impaired children?