Article 2

Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

Empirical Investigation of Visual-Inspection Versus

Trend-Line Analysis of Single-Subject Data


Margaret A Hojem and Kenneth J Ottenbacher
PHYS THER. 1988; 68:983-988.

The online version of this article, along with updated information and services, can
be found online at: http://ptjournal.apta.org/content/68/6/983

Collections This article, along with others on similar topics, appears


in the following collection(s):
Clinical Decision Making
Research: Other
e-Letters To submit an e-Letter on this article, click here or click on
"Submit a response" in the right-hand menu under
"Responses" in the online version of this article.
E-mail alerts Sign up here to receive free e-mail alerts

Downloaded from http://ptjournal.apta.org/ by guest on February 11, 2014


Empirical Investigation of Visual-Inspection Versus
Trend-Line Analysis of Single-Subject Data

MARGARET A. HOJEM
and KENNETH J. OTTENBACHER

We examined the inferential decisions made using either visual analysis alone
or in combination with a trend line to evaluate data from single-subject research
designs. Thirty-nine subjects were randomly assigned to either a Visual Group
(n = 20) that used a visual-inspection approach to analyzing graphed data or a
Quantitative Group (n = 19) that used a trend-line approach. After instruction in
interpretation, we asked the subjects to analyze graphs containing data from five
hypothetical AB single-subject designs. Results revealed a statistically significant
difference in the decision made between the two groups for four of the five
graphs. The group using the trend line to analyze graphed data exhibited more
confidence in the decisions they made and also demonstrated greater within-
group consistency as compared with the group using visual inspection. The
implications of various methods of data analysis in establishing the scientific
legitimacy of single-subject research methods are discussed, and the argument
is made that quantitative procedures can assist in the analysis and interpretation
of single-subject data.
Key Words: Clinical protocols, Physical therapy, Research, Research design.

Researchers in physical therapy have recently reported sev- nesses of single-subject methods before the designs can be
eral investigations using single-subject research designs to appropriately implemented and interpreted.
address questions of clinical and therapeutic interest.1-3 These One area of particular controversy in implementing and
methods, also referred to as single-case experimental, intra- interpreting single-subject designs is data analysis.10 Several
subject, single-system, or idiographic research designs, repre- analytic options are available to the therapist using single-
sent alternatives to traditional group-comparison procedures subject research procedures. In their article addressing the
commonly used in the evaluation of clinical interventions.4,5 issue of interpreting the results of single-subject research de-
In clinical settings, a primary advantage of single-subject signs, Wolery and Harris identify two strategies for analyzing
procedures is their compatibility with clinical practice, which single-subject data: 1) visual analysis of graphed data and 2)
generally focuses on one patient at a time. Other advantages quantitative (statistical) analysis.11
include an emphasis on individualized programs of treatment Graphic presentation and visual analysis have been the
and outcome measures and a focus on understanding the traditional methods used to interpret and draw inferences
process of change through repeated measurements versus from data collected using a single-subject framework.6 Ac-
focusing only on the product or posttest results.6 The com- cording to Kazdin, visual analysis "refers to reaching a judg-
patibility of single-subject designs with clinical practice has ment about the reliability or consistency of intervention ef-
led some investigators to advocate them as the preferred fects by visually examining graphed data."4(p232) Tawney and
method of clinical research in field settings.7,8 Single-subject Gast argue that graphic presentation and visual inspection of
research methods are practice based and practitioner oriented; single-subject data provide practitioners with a compact and
however, they are not an empirical panacea. Connolly et al detailed account of patient performance, including a sequence
have correctly advised physical therapists that single-subject of the design phases, an indication of the time spent in each
methods have advantages and limitations and that therapists phase, and the relationship between the treatment and out-
must understand and appreciate both the strengths and weak- come variable.12
One frequent criticism of visual analysis is the lack of any
formal set of decision rules on which to make inferences from
single-subject data.7 Several authorities have argued that the
M. Hojem, MS, is Physical Therapist, DeForest Public Schools, DeForest,
WI 53532. She was a student in the graduate program in therapeutic science, lack of formal decision rules for making visual judgments
University of Wisconsin-Madison, Madison, WI, when this study was con- from graphed data introduces the possibility of bias and
ducted. subjectivity into the analytical process.13,14 As Kazdin has
K. Ottenbacher, PhD, is Associate Professor, School of Allied Health Profes-
sions, University of Wisconsin-Madison, 2120 Medical Sciences Center, 1300 noted, "the process of visual inspection would seem to permit,
University Ave, Madison, WI 53706 (USA). if not actively encourage, subjectivity and inconsistency in
Address correspondence to Dr. Ottenbacher.
This study was completed in partial fulfillment of the requirements for Ms.
the evaluation of intervention effects."4(p239)
Hojem's master's degree in therapeutic science, University of Wisconsin-Mad- Evidence of inconsistent visual analysis of single-subject
ison. data has been provided by numerous investigators.13-15 Jones
This article was submitted June 8, 1987; was with the authors for revision
four weeks; and was accepted October 21, 1987. Potential Conflict of Inter- et al, for example, compared the use of visual inference and
est: 4. a statistical procedure and found frequent intrarater disagree-

Volume 68 / Number 6, June 1988 983


Downloaded from http://ptjournal.apta.org/ by guest on February 11, 2014
ment and low interrater agreement for visual analysis.16 In a
recent investigation, Ottenbacher found a poor rate of agree-
ment (<.65) for a set of five graphs visually analyzed by 46
occupational therapists.15 When he compared the results of a
visual analysis with those derived from a statistical approach,
he found substantial disagreement on three of the five
graphs.15 The demonstration of poor consistency in the visual
analysis of single-subject data has led several authorities to
advocate the use of some form of quantitative analysis to
supplement the visual inspection of single-subject data.5,7,11
The purpose of this study was to compare the results of
visual versus quantitative analysis of single-subject data under
controlled conditions. We hypothesized that a significant
difference in interpretation would exist between the two meth-
ods and that the confidence in judgments made by individuals
using a quantitative procedure would be greater than that
exhibited by individuals using visual analysis alone.
METHOD
Fig. 1. Single-subject graph (graph 2 from Fig. 2) including example
Subjects of trend line computed using split-middle method of trend estimation.
Thirty-nine upper division students in a school of allied
health professions at a major midwestern university were then projects the trend line into the treatment phase of the
recruited to participate in the study. The subjects (29 women, graph to predict performance in the absence of change. If a
10 men) ranged in age from 21 to 37 years. All subjects signed change occurs in patient performance, the proportion of
an informed consent statement indicating that their partici- observations (data points) above or below the extended line
pation in the study was voluntary. The subjects had completed will be different in the treatment phase than during the
an introductory research methods course but had received no baseline phase. An example of the split-middle method com-
previous training in the visual or quantitative analysis of puted for one of the stimulus graphs used in this study is
single-subject data before participation in the study. shown in Figure 1.
The specific steps used to compute a trend line using the
Procedure split-middle method have been presented extensively in the
single-subject literature4-7 and are not repeated here. Wolery
The 39 students were randomly assigned to either a Visual and Harris have described the computational details and
Group (n = 20) or a Quantitative Group (n = 19). Students presented examples relevant to physical therapists in a pre-
in each group were presented a prepared lesson designed to vious article.11
introduce them to single-subject designs and to teach them Both the visual-analysis and trend-line analysis lessons con-
the components of either visual analysis or the quantitative tained specific objectives. The lessons were delivered to the
trend-line procedure. subjects in a lecture format supplemented with audiovisual
Visual lesson. The visual lesson included a discussion of aids. The lesson plan and specific objectives for each proce-
graphic conventions and examples of the following compo- dure may be obtained from the authors.
nents of visual analysis: level, trend, slope, and variability. Following the respective lessons, each student was requested
Level was defined as an abrupt change that occurs between to interpret a sample graph. During this period of time, any
the last data point in the baseline (A) phase of a research questions concerning the graph, single-subject designs, or the
design and the first data point in the treatment (B) phase of process of interpretation were addressed. In addition, we
the design. Trend referred to the direction indicated in the set requested the Visual Group to complete a brief multiple-
of data points. Trends were defined as stationary, accelerating, choice test to determine whether they had a working knowl-
decelerating, or curvilinear. The visual component of varia- edge of the visual concepts of level, trend, slope, and variabil-
bility referred to the amount of spread or fluctuation in a set ity. The Quantitative Group was asked to compute a trend
of data points. The final visual characteristic described was line and return the sample graph to the investigators (M.A.H.
slope, which was defined as the pitch of the set of data. Slope and K.J.O.). Inspection of the sample graph provided infor-
referred to the steepness of the data path. Specific examples mation concerning whether the subjects had correctly learned
of each of these visual characteristics and various combina- the procedure.
tions of them were presented to the subjects. After the completion of the lesson, we presented a series of
Trend-line lesson. The trend-line procedure presented to five graphs to each group of subjects and requested them to
the Quantitative Group was the split-middle method of trend interpret the graphs by deciding whether a significant change
estimation, as developed by White17 and presented by White in performance occurred across the baseline and treatment
and Haring.18 The split-middle method of trend estimation, phases. The five graphs are presented in Figure 2.
also referred to as the celeration-line approach, is a quantita- The subjects were allowed to use any notes they had taken
tive method of analyzing graphed data rather than a formal and any examples of graphs distributed as part of the lesson.
statistical test. A statistical probability, however, can be com- Each subject was given a copy of thefivegraphs to be analyzed
puted from data produced by the split-middle method of on a separate piece of 8.5- by 11-in* bond paper. Subjects
trend analysis.5 were asked to respond to the statement "There is a significant
The split-middle method of trend estimation uses a non-
technical approach to fit a trend line to the baseline data and * 1 in = 2.54 cm.

984 PHYSICAL THERAPY

Downloaded from http://ptjournal.apta.org/ by guest on February 11, 2014


RESEARCH

GRAPH 1 GRAPH 2

GRAPH 3 GRAPH 4

GRAPH 5

THERE IS A SIGNIFICANT CHANGE IN PERFORMANCE ACROSS THE


TWO PHASES.
STRONGLY STRONGLY
DISAGREE AGREE

Fig. 2. Five single-subject graphs analyzed by Visual and Quantitative Groups.

change in performance across the two phases" by marking Data Analysis


their decision on a horizontal scale. The scale was numbered
0 through 5, with 0 indicating strong disagreement and 5 Subject responses were descriptively analyzed using fre-
indicating strong agreement with the statement (Fig. 2). This quency counts and percentages. We used the Mann-Whitney
statement and response scale were located on the bottom of U test to determine whether the two groups differed in their
the page containing each of the graphs. All subjects were judgments for each of the five graphs. We used the Bartlett
supplied with pencils. In addition, the subjects in the Quan- test to determine whether a significant degree of serial de-
titative Group were supplied with 6-in plastic rulers and a pendency, or autocorrelation, existed within the data. A chi-
copy of Bloom's19 table for determining the statistical signifi- square analysis was performed to evaluate statistically the
cance of any change depicted in the graphs. degree of confidence expressed by the two groups.

Volume 68 / Number 6, June 1988 985


Downloaded from http://ptjournal.apta.org/ by guest on February 11, 2014
RESULTS TABLE 2
The purpose of the lessons received by both the Visual and Median Response Scores and Range for Quantitative and
Visual Groups
the Quantitative Groups was to teach the respective skills
necessary to analyze the graphs. The success of the lesson for Graph Number
the Visual Group was evaluated by the results of the multiple-
1 2 3 4 5
choice test. The Visual Group received an overall score of
81% correct answers on the test. The effectiveness of the Quantitative Group
trend-line analysis lesson was evaluated by comparing the (n = 19)
Median 1 1 5 0 2
celeration line drawn by the subjects on their sample graph
Range 0-5 0-5 4-5 0-3 0-5
to a correct model that we prepared. The sample graphs Visual Group
indicated a 93% correct performance rate for the Quantitative (n = 20)
Group. The scores for both groups indicated that they under­ Median 3 1 1 4 1
stood the analysis components presented during the lesson Range 0-5 0-5 0-3 0-5 0-4
and possessed the basic knowledge necessary to analyze the
graphs.
The frequencies for the values assigned to each of the graphs assumption of independence associated with the use of infer­
by subjects in both the Visual and Quantitative Groups are ential statistical tests. The degree to which this assumption
shown in Table 1. Table 2 presents the median and range for was violated was evaluated by collecting responses from each
each graph for both groups. Table 2 reveals that for three of subject for the five graphs and computing their autocorrela­
the graphs (1,3, and 4), the median scores were located on tion coefficients. The average autocorrelation coefficient for
opposite sides of the midpoint of the scale for the two groups. all 39 subjects was .34 (s = .18). None of the autocorrelation
This finding indicates that a different decision (significant vs values were significant using the Bartlett test.20 Based on this
nonsignificant) was made by the two groups for these three analysis, we considered the subjects' scores to be functionally
graphs. For graphs 1 and 4, the Quantitative Group's median independent and analyzed them accordingly.
score was located on the disagreement end of the scale, The pattern of scores presented in Table 1 suggests an
whereas the Visual Group's median score was located on the uneven distribution of scores for some of the graphs. The
agreement end of the scale. For graph 3, the Visual Group's calculation of third-moment (skewness) coefficients for each
median score was on the disagreement end, whereas the set of scores revealed values greater than 1 for three of the
Quantitative Group's median score was located above the five graphs. Because of the skewed distributions, the numeric
midpoint on the agreement end of the scale. ratings assigned to each graph were compared across the two
One obvious difficulty in analyzing data that are collected conditions using a Mann-Whitney U test. The analysis re­
repeatedly from the same subjects is the possibility that a vealed a statistically significant difference between the ratings
significant degree of serial dependency, or autocorrelation, of the Visual and Quantitative Groups for graphs 1, 3, 4, and
exists within the data. Serial dependency refers to the fact that 5 (p < .05). No statistically significant difference was found
responses provided by the same individual may be correlated. between the two groups for graph 2 (p = .49).
The higher the correlation, the higher the predictive validity To explore whether the Quantitative Group displayed more
of future responses. The result of serial dependency in a data confidence in the decisions they made than the Visual Group,
set is a reduction in the variability of observed responses. The the number of extreme scores (0 or 5) were tabulated for each
presence of serial dependency in the data may violate the graph. A rating of 0 (strongly disagree) or 5 (strongly agree),
indicating a high degree of confidence in the analytical deci­
sion, was selected 55% of the time by subjects in the Quanti­
TABLE 1 tative Group. Subjects in the Visual Group recorded a rating
Frequency of Responses to Significance Statement of 0 or 5 only 24% of the time (Tab. 1). One factor of the chi-
Graph Number square analysis used to evaluate the subject's degree of confi­
Responsea TOTAL dence involved group membership (Visual Group vs Quanti­
1 2 3 4 5 tative Group), and the other factor reflected the presence or
Quantitative Group absence of an extreme score (ie, a score of 0 or 5 vs a score of
(n = 19) 1, 2, 3, or 4). The chi-square analysis revealed a significant
0 9 8 10 2 29 difference between the two groups, with the Quantitative
1 5 4 7 6 22 Group evidencing more extreme scores than the Visual Group
2 4 2 1 3 10 for graph 1 (χ2 = 4.57, p < .05), graph 2 (χ2 = 3.15, p < .05),
3 1 1 1 1 4
and graph 3 (χ2 = 22.28, p < .05). No statistically significant
4 3 1 3 7
5 1 4
difference was revealed for graphs 4 and 5.
18 23
Visual Group To examine the level of agreement that existed within the
(n = 20) Visual and Quantitative Groups, respectively, their responses
0 1 4 3 1 6 15 were dichotomized. Subjects who rated a graph by selecting a
1 1 9 9 1 7 27 response of 3,4, or 5 were scored as agreeing that a significant
2 3 7 3 2 15 change in performance occurred across the baseline and treat­
3 6 5 1 3 3 18 ment phases. Subjects who rated a graph by selecting a re­
4 6 1 7 2 16 sponse of 0,1, or 2 were scored as indicating that no significant
5 3 1 5 9 change occurred across the phases. The percentages of agree­
a
0 corresponds to strongly disagree; 5 corresponds to strongly ment and disagreement for each graph for the Visual and
agree. Quantitative Groups are shown in Table 3.

986 PHYSICAL THERAPY

Downloaded from http://ptjournal.apta.org/ by guest on February 11, 2014


RESEARCH
A total of 10 statistical tests were computed in these two visual analysis exclusively poses several problems. One of
sets of analyses. The presence of multiple statistical compari­ these problems is the subjective nature of the inference made
sons means that the experimentwise error rate for the study concerning the independent variable.4 Parsonson and Baer
may be inflated beyond the per-comparison level of .05. One believe that only interventions that produce dramatic results
possible solution to this problem is to use a technical adjust­ are worthy of continued consideration as research variables,
ment such as the Bonferroni correction.21 The Bonferroni so the subjectivity of the evaluation of weaker treatment
correction would reduce the per-comparison error rate and effects is not viewed as a problem.23 Kazdin is concerned
thus help control the overall or experimentwise error rate. A about small, but reliable, treatment effects, because he feels
disadvantage of the Bonferroni procedure is that the Type II that ignoring small treatment effects that are consistent may
error rate is increased as the per-comparison alpha level is leave some treatments unexplored that could have a cumu­
reduced. The experimentwise error rate in this study was lative effect or that could be strengthened with further
evaluated using the percentage error rate procedure reported work.4 This controversy can be reduced to the argument over
by Ottenbacher.22 The percentage error rate (PE) indicates the whether small and medium treatment effects can be clinically
proportion of results labeled statistically significant that are significant, that is, have effects that would be important to
likely to occur by chance and is computed by the formula the patient. Cohen contends that small treatment effects are
more likely to be found in new areas of investigation or in
PE = 100α/M applied fields where the phenomena being studied are typi­
where c is the total number of statistical tests conducted, α is cally not under good experimental or measurement control.25
the per-comparison alpha level, and M is the number of Small effects should be common in areas of clinical research
statistical tests significant at the per-comparison alpha level. where the variables under investigation are manifested in
In this study, PE = 100(10).05/7 = 7.1%, indicating that subtle ways and may be difficult to manipulate. It seems
about 1% of the results may have occurred by chance, whereas reasonable to assume that many of the effect sizes of interest
the remaining 93% did not. This finding strongly suggests in physical therapy clinical research would fall into this cate­
that a Type I error was unlikely, despite the fact that multiple gory. If these smaller intervention effects are viewed as poten­
statistical comparisons were conducted. tially important, visual methods of analysis, when used alone,
clearly may not be adequate to detect them.15
DISCUSSION The subjectivity of visual analysis is partially the result of
The results of this investigation revealed a statistically sig­ the lack of clear-cut decision rules for determining whether a
nificant difference between the Quantitative and Visual treatment effect has taken place. Because judgments using
Groups in their rating of four of the five graphs. The Quan­ visual analysis result from a complex function of changes in
titative Group used more extreme scores in rating the graphs slope, level, trend, and variability, it has been difficult to
than the Visual Group, suggesting a higher degree of confi­ arrive at consistent decision rules for determining treatment
dence in their decisions. The within-group level of agreement effects using visual analysis.5,13
was also higher on four of thefivegraphs for the Quantitative In contrast to pure visual inference, analysis of single-
Group, indicating greater agreement or consistency in the subject data using the split-middle method of trend estimation
ratings. appears to be associated with greater consistency and confi­
The two groups demonstrated a difference in the direction dence.16 The split-middle method of trend analysis is relatively
of their responses for three of the five graphs. This result easy to use and can be applied in any clinical setting. It is
suggests that a different interpretation was being made (sig­ computed as an adjunct to visual analysis and requires that
nificant vs nonsignificant) by the two groups. Thisfindinghas the data be graphed and visually interpreted as well. Statistical
obvious implications when comparing the results of single- significance can be determined using this method in combi­
subject research studies analyzed using a purely visual method nation with Bloom's table.19 If the split-middle trend line is
versus a combination of visual and quantitative methods. drawn correctly, the results should be consistent across re­
searchers.
Implications Concerning Clinical Versus The presence of statistical significance, however, does not
Statistical Significance guarantee the clinical significance of the findings. Kazdin4
and Bloom and Fischer5 recommend the use of social vali­
The use of visual versus quantitative methods of analysis dation techniques in addition to visual and quantitative analy­
for single-subject studies is controversial.23,24 Reliance on sis to help address questions of clinical relevance. Social
validation refers to consideration of socially important criteria
TABLE 3 for evaluating the treatment and its clinical significance. This
Percentages of Agreement and Disagreement for Quantitative technique can be applied through comparison of the patient's
and Visual Groups performance to normative data or through observations by
Graph Number significant individuals in the patient's everyday life.4
1 2 3 4 5
Limitations
Quantitative Group
(n = 19) When assessing the implications of these findings, the fol­
Agreement (%) 95 74 100 95 42 lowing limitations should be considered. Both groups received
Disagreement (%) 5 26 0 5 58 minimal training in the analysis method used. The training
Visual Group received was not enough to completely and thoroughly master
(n = 20)
either of the methods. The rates of agreement for the Visual
Agreement (%) 75 65 95 75 75
Disagreement (%) 25 35 5 25 25
Group, however, were similar to those made by "experts" in
visual analysis in other studies.13,14 It seems unlikely, there-

Volume 68 / Number 6, June 1988 987


Downloaded from http://ptjournal.apta.org/ by guest on February 11, 2014
fore, that within-subject agreement could be significantly scientific research methods.26 He also maintains that the
improved in the visual-analysis condition. The Quantitative clinical therapist is in the best position to identify appropriate
Group possibly might reach higher levels of agreement with questions to be asked through research. Single-subject meth-
further practice. It is also important to note that the subjects ods have many advantages that would make them useful for
were students with limited experience in the systematic ob- evaluating clinical change in physical therapy.7 Controversies
servation and analysis of patient performance. concerning the analysis of data from single-subject research
Another limitation relates to the number of data points in designs must be addressed, however, before these designs can
each phase of the graphs. Eight observations is the lower limit be accepted as scientifically legitimate.
recommended to compute the split-middle method of trend Although quantitative methods may be redundant with
estimation.18 A larger number of data points would improve visual inference in the presence of large, clearly defined treat-
the accuracy of the celeration line. Smaller data sets, however, ment effects, the use of a quantitative adjunct apparently can
comparable to those used in this study, are more representa- increase consensus and lend a greater degree of confidence to
tive of current practices in single-subject research.10,21 the analytical decision when the results are equivocal or the
The unit of analysis used in this study (the AB design) also treatment effects are small. Because the split-middle method
may represent a limitation. The AB design is the basic unit of trend estimation requires that the data be graphed, it forces
of analysis in single-subject research. The AB configuration the investigator to combine both visual and quantitative
serves as the foundation for more complex multiple baseline procedures. This combination of methods is an advantage
and reversal designs. In practice, however, the simple AB over some other statistical procedures that can be computed
design is rarely used as the functional framework to manipu- without graphing and visually inspecting the data.5
late the independent variable. More complex designs, includ- A great deal of additional research is required before an
ing those with several phase changes or multiple baselines, empirical consensus can be reached concerning the appropri-
may have produced different results. ate methods of data analysis for single-subject research de-
Finally, the graphs did not represent actual data. They were signs. Future studies should address the levels of agreement
constructed to represent specific properties that might be between various quantitative methods or the effect of an
related to visual analysis, but are only a small sample of the increased number of phase changes on the interrater agree-
possible situations that might appear in genuine data. It is ment of visual analysis. The studies conducted so far have
important to remember, therefore, that this is not an exhaus- not explored the effect of the researchers' knowledge of the
tive exploration of agreement between these two methods. variables under consideration on the outcome decision or on
interrater agreement. This is another area requiring further
CONCLUSIONS exploration. The data generated by these and other investi-
Currier notes the need for physical therapists to test knowl- gations will provide therapists with the information needed
edge that has been handed down through tradition using to make valid scientific inferences from single-subject designs.

REFERENCES
1. Laskas CA, Mullen SL, Nelson DL, et al: Enhancement of two motor 15. Ottenbacher KJ: Reliability and accuracy of visually analyzing graphed data
functions of the lower extremity in a child with spastic quadriplegia. Phys from single-subject designs. Am J Occup Ther 40:464-469, 1986
Ther 65:11-16, 1985 16. Jones RR, Weinrott MR, Vaught RS: Effects of serial dependency on the
2. Martin JE, Epstein LH: Evaluating treatment effectiveness in cerebral palsy: agreement between visual and statistical inference. J Appl Behav Anal
Single-subject designs. Phys Ther 56:285-294, 1976 11:277-283,1978
3. Harris SR, Riffle K: Effects of inhibitive ankle-foot orthoses on standing 17. White OR: The "Split-Middle": A "Quickie" Method of Trend Estimation.
balance in a child with cerebral palsy: A single-subject design. Phys Ther Seattle, WA, The University of Washington, Experimental Education Unit,
66:663-667, 1986 Child Development and Mental Retardation Center, 1974
4. Kazdin AE: Single-Case Research Designs: Methods for Clinical and Ap- 18. White OR, Haring NG: Exceptional Teaching: Individually Planned Educa-
plied Settings. New York, NY, Oxford University Press Inc, 1982 tions, ed 2. Columbus, OH, Charles E Merrill Publishing Co, 1980
5. Bloom M, Fischer J: Evaluating Practice: Guidelines for the Accountable 19. Bloom M: The Paradox of Helping: Introduction to the Philosophy of
Professional. Englewood Cliffs, NJ, Prentice-Hall Inc, 1982 Scientific Practice. New York, NY, John Wiley & Sons Inc, 1975
6. Barlow D, Hersen M: Single-Case Experimental Designs: Strategies for 20. Glass GV, Willson VL, Gottman JM: Design and Analysis of Time-Series
Studying Behavior Change, ed 2. Elmsford, NY, Pergamon Press Inc, 1984 Experiments. Boulder, CO, Colorado Associated University Press, 1975
7. Ottenbacher KJ: Evaluating Clinical Change: Strategies for Occupational 21. Harris RJ: A Primer of Multivariate Statistics. New York, NY, Academic
and Physical Therapists. Baltimore, MD, Williams & Wilkins, 1986 Press Inc, 1975
8. Kazdin AE: Single-case research in clinical child psychiatry. J Am Acad 22. Ottenbacher KJ: A quantitative analysis of experimentwise error rates in
Child Psychiatry 22:423-432, 1983 applied behavioral science research. Journal of Applied Behavioral Science
9. Connolly BH, Craik RL, Krebs DE: Guest editorial: Single-case research- 22:495-501,1986
When is it valid? Phys Ther 63:1767-1768, 1983 23. Parsonson BS, Baer DM: The analysis and presentation of graphic data.
10. Kratochwill TR, Brody GH: Single-subject designs: A perspective on the In Kratochwill TR (ed): Single-Subject Research: Strategies for Evaluating
controversy over employing statistical inference and implications for re- Change. New York, NY, Academic Press Inc, 1978, pp 101-165
search and training in behavior modification. Behav Modif 2:291 -307, 1978 24. Parsonson BS, Baer DM: The graphic analysis of data. In Poling A, Fuqua
11. Wolery M, Harris SR: Interpreting results of single-subject research de- W (eds): Research Methods in Applied Behavior Analysis. New York, NY,
signs. Phys Ther 62:445-452, 1982 Plenum Publishing Corp, 1986, pp 157-186
12. Tawney JW, Gast DL: Single-Subject Research in Special Education. 25. Cohen J: Statistical Power Analysis in the Behavioral Sciences, rev ed.
Columbus, OH, Charles E Merrill Publishing Co, 1984 New York, NY, Academic Press Inc, 1977
13. DeProspero A, Cohen S: Inconsistent visual analysis of intrasubject data. 26. Currier DP: Elements of Research in Physical Therapy, ed 2. Baltimore,
J Appl Behav Anal 12:573-599, 1979 MD, Williams & Wilkins, 1984
14. Furlong MJ, Wampold BE: Intervention effects and relative variation as
dimensions in experts' use of visual inference. J Appl Behav Anal 15:415-
421,1982

988 PHYSICAL THERAPY


Downloaded from http://ptjournal.apta.org/ by guest on February 11, 2014
Empirical Investigation of Visual-Inspection Versus
Trend-Line Analysis of Single-Subject Data
Margaret A Hojem and Kenneth J Ottenbacher
PHYS THER. 1988; 68:983-988.

Cited by This article has been cited by 2 HighWire-hosted articles:

http://ptjournal.apta.org/content/68/6/983#otherarticles
Subscription http://ptjournal.apta.org/subscriptions/
Information
Permissions and Reprints http://ptjournal.apta.org/site/misc/terms.xhtml
Information for Authors http://ptjournal.apta.org/site/misc/ifora.xhtml

Downloaded from http://ptjournal.apta.org/ by guest on February 11, 2014

You might also like