Professional Documents
Culture Documents
A Graph Paints 1,000 Words But are they Valid?: A Systematic Review of SCRD
Graphs
1Cian L. Brown, Corey Peltier, David Y. Lee, Fanee R. Webster, Amal A. Shabibi
1Department of Educational Psychology, University of Oklahoma
Author Note
Abstract
i. Objective: Single-case research designs are a useful methodology to investigate the effects of
treatments within counseling. A core component of the design is the presentation of data
collected repeatedly on a time-series graph. Research in the fields of special education and
behavior analysis suggests most graphs do not adhere to reporting guidelines and specific
iii. Results: Most of the graphs did not meet recommended guidelines for graph construction.
iv. Conclusions: We provide a discussion on how graph construction may impact visual analysis
The use of single-case research designs (SCRDs) is increasing in counseling research and
practice to establish evidence for treatment and intervention effectiveness. Although SCRD has
Traditionally, counseling research has relied upon research designs using parametric tests to
determine statistical significance. These designs require large sample sizes, usually a
applicability towards implementing these research approaches have led to researchers to explore
alternative modalities using smaller samples and consider clinical significance (Lenz, 2015).
Johnson and Cook (2019) highlight SCRDs use for both confirmatory and exploratory analyses,
therefore researchers can examine existing practices and/or explore new interventions impact
on behavioral and mental health outcomes using a single subject (i.e., individuals, groups,
families). Therefore, interventions in SCRD focus primarily on what works for individual clients,
initial assessed outcomes or treatment (Mirza et al., 2017; Scuffham, et al., 2010). Thus, SCRDs
provide clinicians and researchers with a viable means to identify and explore agents of change
while measuring experimental outcomes (without confronting the dauting demands of achieving
statistical power).
A unique feature of SCRDs is the researcher’s ability to measure client outcomes multiple
as their own control, eliminating the need for control groups usually required for between-
subject designs. Once baseline has been established, the intervention or treatment condition is
introduced, the researcher continues to measure the dependent variable over several timepoints
to assess change in the client’s outcome. This design characteristic of SCRD highlights the
unresponsive to the intervention, the researcher has flexibility to modify the treatment.
Typically, treatment effectiveness is determined by evaluating the level, trend, and variability
within each phase of data collection and consider the level change, immediacy of effect, and
Prior to 2015, counseling research using SCRD was sparse. Woo et al. (2016) explored
the use of SCRDs in counseling research across the 20 American Counseling Association (ACA)
journals from 2003 to 2014, which yielded 7 studies across 5 journals. Notably, in 2015, the
Journal of Counseling & Development released a special issue highlighting SCRD and its
applicability within the counseling field. Since then, counseling researchers have employed
SCRDs to their work reinforcing SCRD as a practical application to use across clinical and
research settings to examine the effectiveness of treatment without having to employ non-
treatment groups or conduct large sample parametric tests. Therefore, this methodolody
substantially diverges from other design options (i.e., group design approaches) through the
collection of time series data and visual analysis of the graphed data to illustrate changes and
treatment effectivness, hallmark characteristics of SCRD. Recent work has found large
al., 2017; Kubina et al., 2022; Ledford, et al., 2019; Peltier et al., 2021; Peltier et al., 2022a;
There are a host of decisions that must be made when constructing a time-series graph.
Dart and Radley (2018) provided a framework for researchers and clinicians to consider
classifying graphical elements as either aesthetic altering or analysis altering. Aesthetic altering
elements are those that change the way a graph looks but to date do not have evidence
suggesting when manipulated they change the decisions made by a person evaluating the graph.
SCRD Graph Construction in Counseling 5
Much like APA style for writing and reporting results in manuscripts, standardizing aesthetic
altering elements enhances graph reading. Analysis altering elements are those with evidence
that when manipulated will alter the decisions made by a person evaluating the graph. Thus,
particular attention must be paid to analysis altering elements because these impact conclusions
There are two potential analysis altering elements for SCRD time-series graphs –
ordinate scaling and the DPPXYR. For ordinate scaling, Dart and Radley (2017) found that when
graphs displayed a truncated ordinate (i.e., setting y-max to 80%, 60%, or 40%) Type I errors
increased based on the severity of the truncation. The DPPXYR is a metric that captures the
ratio of the x:y axis lengths while also considering the density of the data points plotted along
the x-axis – it is computed using the following formula (x-axis length/y-axis length)/number of
data points that could be plotted along the x-axis. Radley and colleagues (2018) suggested an
ideal ratio of DPPXYR between 0.14 and 0.16, graphs with DPPXYR less than 0.14 led to inflated
Type I errors. We highlight these are potentially analysis altering because more replications are
needed; in a conceptual replication Peltier et al. (2022a) found no evidence for ordinate scaling
or DPPXYR impacting visual analyst decisions. We refer readers to Figure 1 to see examples of
Figure 1
published in the counseling literature – hence the purpose for the current project. However,
there have been several reviews of graph construction in the fields of behavior analysis and
special education.
reported in SCRDs across 11 behavior analytic journals from each journal's inception through
2011. The authors identified several characteristics that were inconsistent with current
recommendations: (a) failing to label the x- and y-axis, (b) misuse of tick marks on the x- and y-
axis, (c) inconsistent scaling of the axis, and (d) ratio of x-axis length to y-axis height did not
adhere to the recommended ratio (i.e., between 8:5 and 4:3). Kubina and colleagues (2021)
education journals. Results paralleled their review of behavior analytic journals, and the authors
highlighted the need to train and disseminate guidelines for time-series graph construction.
A separate research team conducted a similar project. Ledford and colleagues (2019)
education journals. Like Kubina and colleagues (2017; 2021), the authors found large variation
in x:y axis ratio – both across types of SCRDs (e.g., multiple-baseline design vs. ABAB) and
within each design type. Ledford and colleagues (2019) aimed to investigate what may be
influencing varied graph construction and the lack of adherence to recommended guidelines.
They surveyed 50 editorial board members across the journals to determine if they preferred
graphs with an x:y ratio that (a) fit recommended guidelines, (b) was most prevalent in
published graphs, or (c) neither. Most respondents reported preferences for x:y axis that were
divergent from recommended guidelines and what was most prevalent in the published
literature.
SCRD Graph Construction in Counseling 8
Three reviews (Peltier et al., 2021; Peltier et al., 2022a; Peltier et al., 2022b) extended
prior reviews by focusing on the two potentially analysis altering elements: (a) ordinate
truncation (see above) and (b) DPPXYR (see above). Each review focused on evaluating a
decade’s worth of time-series graphs reporting SCRDs in specific subfields of special education:
Peltier and colleagues (2021) evaluated 315 graphs in the learning disabilities journals; Peltier et
al. (2022a) evaluated 258 graphs in the emotional or behavioral disorders journals; and Peltier
et al. (2022b) evaluated 2,675 graphs in autism journals. Findings across reviews reached
similar conclusions. The authors found few graphs displayed a truncated ordinate, which is
promising because ordinate truncating was shown to increase Type I errors. However, the x:y
axis ratio and DPPXYR varied dramatically across graphs with few meeting current
reccomendations.
Purpose
Graph construction impacts readability of graphs and may distort the visual analysis
process for SCRDs. Prior reviews have provided evidence and actionable suggestions to improve
graph construction in behavior analysis and special education. To date, there has not been a
review of time-series graphs in counseling, so little is known on the variability and level of
adherence to recommendations for time-series graph construction. The purpose for this review
We hypothesize less than 10% of time-series graphs will have a DPPXYR value within the
recommended range (i.e., 0.14 to 0.16), and less than 10% of time-series graphs will have a x:y
axis ratio within the recommended range (i.e., 8:5 to 4:3). We will also explore additional graph
Method
journals to provide descriptive data related to the graphical characteristics of SCRDs conducted
in counseling.
Search Procedure
Sample Selection
Like Woo et al. (2016), we selected ACA affiliated journals (n = 20) to include in the review.
Educators and Counseling practitioners and considering journals associated with regional
associations and professional organizations common among licensed counselors. This resulted
in 42 journals being included in the review. Journal issues were selected and screened between
the first issue published in 2015 to the last issue published in 2021. This period was selected
based upon the release of the special issue in Journal of Counseling & Development (see above)
and extended upon the SCRD content analysis review conducted by Woo et al. (2016) - they
searched from 2003 through 2014. Access the preregistered document link for included journal
list (https://doi.org/10.17605/OSF.IO/FU7GV).
To be included studies needed to meet the following criteria: (1) published or assigned to
an issue in years 2015 to 2021; (2) published in one of the specified journals above; (3) used a
SCRD; and (4) included a time-series graph. Articles were excluded if they (a) were published
before 2015 or after 2021; (b) were published in a journal outside the ones specified above; (c)
used a group design, case study design, or any other type of methodology not fitting a single-
Screening Procedures
Two graduate students completed screening activities. The lead authors provided a
training consisting of the following activities: (a) providing a clear definition for each inclusion
and exclusion criterion, (b) providing examples and non-examples for each inclusion criterion,
(c) providing opportunities for questions and clarifications, and (d) practice trials for a sample
of articles. Following the training, the graduate students screened 6,950 documents by
The next stage consisted of full text screening. The same graduate students screened 118
full texts against the inclusion. We identified 50 studies that met inclusion criteria (see Figure 2
Figure 2
Note. JCD = Journal of Counseling and Development; IJPT = International Journal of Play
Therapy; JCAC = Journal of Child & Adolescent Counseling; CORE = Counseling Outcome Research
& Evaluation; JSC = Journal of School Counseling; JSGW = Journal for Specialists in Group Work;
JCMH = Journal of Creativity in Mental Health; PSC = Professional School Counseling; TPC = The
Professional Counselor; IJAC = International Journal for the Advancement of Counselling; JAOC =
Journal of Addictions & Offender Counseling; JCLA = Journal of Counselor Leadership &
Advocacy; JCRP = Journal of Counseling Research & Practice; JHC = Journal of Humanistic
Counseling; JMHC = Journal of Mental Health Counseling
SCRD Graph Construction in Counseling 12
screening decisions, 12 of the 42 journals (i.e., 28.6% journals) including 2,359 documents
(34.5% of total documents) were independently screened by both graduate students. We used
agreement (IRA). An agreement consisted of either (a) both students excluding the study or (b)
both students retaining the study for full text screening. Mean IRA across journals was 99.6%
(SD = 0.5%, Range = 98% to 100%). The 12 disagreements were resolved by the lead authors
Data Extraction
We consulted previous research (Kubina et al., 2017; Ledford et al., 2019; Peltier et al.,
2021; Peltier et al., 2022a; Peltier et al., 2022b) to develop a coding guide. The coding guide
characteristics, and graph characteristics. The pre-registered coding guide included four
variables related to source information, nine variables related to experimental design, and 21
variables related to graph characteristics. Through coding activities, we identified five additional
variables that we added to the coding guide. These variables were not included in the pre-
registration and thus were not decided a priori. Please refer to the following link for operational
The research team consisted of two faculty members who held a doctoral degree and had
experience conducting SCRDs and three graduate students with coursework related to SCRDs.
For the training, a coding guide was developed with each coded variable identified and a
definition to support accurate coding. The team discussed each variable and identified examples
and non-examples from published graphs. Each coder independently coded two graphs and
agreements and disagreements were discussed. Coders were above 90% accurate on the training
IRA for coding was estimated by comparing exact agreement for each coded variable
independently double coded 17 of 50 (34%) studies including 81 of 272 graphs (29.8%). The
second team of coders independently double coded 49 of 272 graphs (18%) that were randomly
selected from 29 of 50 studies (58%). In total, we double coded 130 of 272 graphs (47.8%)
printed in 46 of 50 studies (92%). For the first team of coders, mean IRA at the variable level
was 89.3% (SD = 18.7%, Range = 30% to 100%) and mean IRA at the graph level was 89.3% (SD
= 4.4%; Range = 78.9% to 100%). For the second team of coders, mean IRA at the variable level
was 94% (SD = 11.2%, Range = 53% to 100%) and mean IRA at the graph level was 94% (SD =
4.9%; Range = 78.9% to 100%). For the first team of coders the two variables with low IRA
included ordinate truncation and whether the graph included a break in the data path across the
phase change line. For the second team of coders the variable with the lowest IRA was whether
the graph included labels for each phase. All disagreements were resolved by the two lead
authors.
Data Analysis
Once articles meeting criteria were screened and selected for inclusion, we examined all
graphs with a vertical and horizontal axis with longitudinal data. The graph must have contained
a maximum one data point per series on the horizontal axis interval excluding scatterplot graphs
corresponding to the subject. A unit of time or sessions must be present on the horizontal axis
along with a quantitative value on the vertical axis. To answer our research question, we
examined each included graph individually whether appearing alone or in the context of other
graphs. Each graph was scored for the essential structure of y and x axis, which is concerned
with representation of quantity and time (e.g., axis labels, min and max value range, ratio, unit
of time, sessions, truncation, etc.), and the presence or absence of time-series graph quality
features (e.g., axis label markers, tick marks present, tick mark style, scaling distance, data
points, data paths, condition change lines, condition break changes, condition labels, data
SCRD Graph Construction in Counseling 14
aligned with tick marks, figure captions), and common graph characteristics for SCRD (e.g., y-
axis floated, data on y-axis, horizontal lines present, equal session intervals, number of DVs
present, number of participants present) (See Kubina et al., 2017; Table 1). Additionally, we
examined the DPPXYR and standardized y:x ratio by using the formula (x-axis length/y-axis
length)/number of data points that could be plotted along the x-axis, to obtain the plot density.
Results
Results of our systematic search yielded a total of 272 graphs reported in 50 articles
printed in 12 counseling journals. Within each article, the median number of graphs included
was 4 and the mean number of graphs included was 4.66 (SD = 3.50; Range = 1-15).
Essential Structure
One essential structure of a time-series graph involves an ordinate axis (i.e., vertical) and
abscissas axis (i.e., horizontal). All graphs evaluated included these axes. Two other essential
features for a graph involve meaningful labels for the ordinate and abscissas axis so viewers
know the nature of the dependent variable and the time scale displayed along the abscissas. We
identified 43 graphs (15.8%) did not provide a label for the ordinate axis and 6 graphs (2.2%)
did not provide a label for the abscissas. See Table 1 for results.
Table 1
Quality Features
We coded seven features related to the ordinate axis. Two variables were specifically
related to the use of tick marks, and these had high rates of divergence from recommendations:
(a) no use of tick marks (45.2% error) and (b) style of tick mark used was not outside graph
space (57.35% error). We coded two variables related to the scaling of the ordinate axis, all
graphs used equal spacing when scaling the axis (0% error) but few graphs floated the ordinate
axis to prevent minimum values from sitting on the abscissas axis (98.2% error). Last, we coded
whether data points fell on the ordinate axis (0% error) and whether the graph included
We coded five variables related to the abscissas axis. Three variables were specifically
related to the use of tick marks, and these had high error rates: (a) no use of tick marks (52.6%
error), (b) style of tick mark used was not outside graph space (45.6% error), and data points did
not align vertically with the tick mark on the abscissas (72.05% error). Two variables related to
scaling the abscissas, (a) the scaling distance was not equal interval along the abscissas (24.6%
error) and (b) session labels were not consistent with time unit of measurement systems (17.6%
error).
We coded five variables related to the data points and paths for each phase of the
experiment. We identified few errors for three variables, (a) data points were not present on the
graph (5.9% error), (b) data paths were not visible or provided for each condition (6.6% error),
and graph did not include vertical phase change lines to signal alterations in the experiments
phase (8.45% error). Two features that included higher error rates included (a) not providing a
break in the data path across phase change lines (40.8% error) and (b) failing to label conditions
Table 2
Abscissas Axis
Tick Marks Present No: 143 52.60%
Yes: 129
Tick Mark Style None: 143 45.60%
Inside: 5
Cross: 0
Mixed: 0
Outside: 124
Data Align Tick Mark No: 196 72.05%
Yes: 76
Data Paths
Data Points Present No: 16 5.90%
Yes: 256
Data Paths No: 18 6.60%
Yes: 254
Condition Change Line No: 23 8.45%
Yes: 249
Break Condition Change No: 111 40.80%
SCRD Graph Construction in Counseling 17
Yes: 161
Condition Labels No: 85 31.25%
Yes: 187
Other
Number of DVs Present 1: 233
2:10
3:19
4:08
5:00
6:02
Axes Comparison
We coded two other elements related to the number of data paths presented on the
graph. First, we coded the number of dependent variables presented on the individual graph
space. Most graphs included one dependent variable (n = 233, 85.7%) although we did identify
graphs that displayed two dependent variables (n = 10), three dependent variables (n = 19), four
dependent variables (n = 8), and six dependent variables (n = 2). Second, we coded the number
of participants data presented on the individual graph space. Most graphs included one
participant’s data (n = 239, 87.9%) although we did identify graphs that displayed two
participants’ data (n = 12), three participants’ data (n = 12), four participants’ data (n = 5), five
participants’ data (n = 3), and 11 participants’ data (n = 1).We coded two variables that have
potential evidence suggesting they are analysis altering elements (i.e., ordinate truncation,
DPPXYR) – and a related element related to the x:y ratio. We identified 178 graphs (65.4%) that
displayed a scale that truncated the minimum and/or maximum ordinate values given the
value less than 0.14 and 162 graphs with a DPPXYR greater than 0.16, thus 83.1% were outside
the recommended guidelines (i.e., 0.14 to 0.16). Last, we coded the standardized x:y ratio, 11
were less than 1.33 and 248 were greater than 1.6, thus 95.2% were outside the recommended
the distribution of standardized x:y values and the DPPXYR. The mean of the standardized x:y
values was 3.025 (95% CI = 2.875, 3.175; SD = 1.26) and the median was 2.56. The distribution
of standardized x:y values was moderately positively skewed (skewness = 0.65) with the tails of
the distribution not accounting for a significant amount of the data set (kurtosis = -0.54). The
mean of DPPXYR values was 0.23 (95% CI = 0.22, 0.25; SD = 0.14) and the median was 0.18.
The distribution of DPPXYR values was moderately positively skewed (skewness = 1.00) with
the tails of the distribution accounting for a significant amount of the data set (kurtosis = 1.51).
See Figure 3 for violin plots displaying the distribution of standardized x:y and DPPXYR values.
Table 3
Figure 3
Violin Plots Displaying the Distribution of Standardized X:Y Ratio and DPPXYR
Discussion
While SCRD research is not new, the field of counseling is using the design at an
increasing rate as evidenced by the number of publications over the past six years. As
characteristics of consistently implementing, analyzing, and reporting the measures and results
of the design. Regarding our hypotheses, less than 10% of the time-series graphs have a
standardized x:y axis ratio within the standardized ratio range (i.e., 8:5 to 4:3), with 95.2% of
graphs falling outside the recommended range. However, our hypothesis that less than 10% of
SCRD Graph Construction in Counseling 20
graphs will have a DPPXYR value within the recommended range (i.e., 0.14 to 0.16) was not
met; 83.1% fell outside the recommended range, meaning 16.9% of reported time-series graphs
met the recommended range. Additionally, 65.4% of the graphs were ordinately truncated,
shortening of the y-axis), which can skew the visual representation of data. These results mean a
majority of the SCRD studies included may have had analysis altering decisions and results
influenced by the visual analysis. Results for standardized x:y parallel prior reviews (Kubina et
al., 2017; 2021; Ledford et al., 2019; Peltier et al., 2021; 2022a; 2022b), which identified more
than 90% of graphs were not within the standardized x:y recommendations. However,
substantially more graphs in the counseling literature displayed a truncated ordinate compared
to prior reviews. This is likely influenced by the measurement systems used, with counseling
construction known as essential structure and quality features. Regarding the ordinate axis,
nearly half of graphs did not present tick marks, of those that did, a little over 43% accurately
presented tick marks outside the axis. A large percent of the graphs are truncated, do not float
the y-axis, do not align data with tick marks, and display horizontal lines through the graph. For
the abscissas axis, similar percentages were found regarding tick marks being present and
properly displayed outside the axis. Notably, almost a quarter of the graphs did not use proper
scaling distance between session/time intervals. Regarding data paths, most graphs presented
data points and paths correctly with condition change lines present. However, 30% to 40% of
graphs did not display condition labels or present a break in the condition change.
Implications
Because the visual analysis process and graph readability are imperative to SCRD
standards is important to informing future research. The reported graph characteristics results
SCRD Graph Construction in Counseling 21
indicate counseling researchers using SCRD methodologies can benefit from a supportive guide
for constructing and presenting SCRD graphs. This supportive guide outlining adherence to
graph construction standards will also be a valuable resource for future readers and reviewers
when visually inspecting time-series graphs. Additionally, journals may want to provide editorial
guidance or additional information on graph construction standards beyond file type and
resolution. Future research may consider performing a content analysis to better understand the
information presented in SCRD graphs and be used to inform procedural guidelines for SCRD
graph construction related to the counseling field. For instance, Ledford et al. (2019) argues for
consistent representation of time on the x-axis since the term ‘sessions’ is not an ordinal
representation of time.
Conclusion
The present study explored the descriptive information of 272 time-series graph
construction impacts readability of graphs and may distort the visual analysis process for
SCRDs. The results indicated several areas of improvement based on errors in graph
construction and visual display of data for counseling researchers and practitioners. For
instance, the greatest area of improvement would be the inclusion and use of tick marks and
data aligned, floating the y-axis, consistent y-scaling ranges to avoid truncation, and
considerable emphasis on scaling the axes to recommended DPPXYR and standardized x:y
ratios.
SCRD Graph Construction in Counseling 22
References
Dart, E. H., & Radley, K. C. (2017). The impact of ordinate scaling on the visual analysis of
https://doi.org/10.1016/j.jsp.2017.03.008
Dart, E. H., & Radley, K. C. (2018). Toward a standard assembly of linear graphs. School
Johnson, A. H., & Cook, B. G. (2019). Preregistration in single-case design research. Exceptional
Kubina, R. M., Kostewicz, D. E., Brennan, K. M., & King, S. A. (2017). A critical review of line
https://doi.org/10.1007/s10648-015-9339-x
Kubina, R. M., Jr., Kostewicz, D. E., King, S. A., Brennan, K. M., Wertalik, J., Rizzo, K., &
review of their use and relevance. Education & Treatment of Children, 44(4), 275–290.
https://doi.org/10.1007/s43494-021-00053-3
Ledford, J., Lane, J., & Severini, K. (2018). Systematic use of visual analysis for assessing
https://doi.org10.1017/BrImp.2017.16
Ledford, J. R., Barton, E. E., Severini, K. E., Zimmerman, K. N., & Pokorski, E. A. (2019). Visual
display of graphic data in single case design studies: Systematic review and expert
Lenz, A. S. (2015). Using single-case research designs to demonstrate evidence for counseling
https://doi.org/10.1002/jcad.12036
SCRD Graph Construction in Counseling 23
Mirza, R. D., Punja, S., Vohra, S., & Guyatt, G. (2017). The history and development of N-of-1
https://doi.org/10.1177/0141076817721131
Peltier, C., Morano, S., Shin, M., Stevenson, N., & McKenna, J. W. (2021). A decade review of
Peltier, C., McKenna, J. W., Sinclair, T. E., Garwood, J., & Vannest, K. J. (2022a). Brief report:
Ordinate scaling and axis proportions of single-case graphs in two prominent EBD
https://doi.org/10.1177/0198742920982587
Peltier, C., Muharib, R., Haas, A., & Dowdy, A. (2022b). A decade review of two potential
Radley, K. C., Dart, E. H., & Wright, S. J. (2018). The effect of data points per x-to y-axis ratio on
314-322. https://psycnet.apa.org/doi/10.1037/spq0000243
Scuffham, P. A., Nikles, J., Mitchell, G. K., Yelland, M. J., Vine, N., Poulos, C. J., ... & Glasziou,
P. (2010). Using N-of-1 trials to improve patient management and save costs. Journal of
Woo, H., Lu, J., Kuo, P., & Choi, N. (2016). A content analysis of articles focusing on single-case
research design: ACA journals between 2003 and 2014. Asia Pacific Journal of
https://doi.org/10.1080/21507686.2016.1199439