Behavioral Science for Data Scientists: Why Your Dashboard Users Think Differently Than You Do

The hardware and bandwidth for this mirror is donated by METANET, the Webhosting and Full Service-Cloud Provider.
If you wish to report a bug, or if you are interested in having us mirror your free-software or open-source project, please feel free to contact us at mirror[@]metanet.ch.

Introduction: The Gap Between Data Creators and Data Users

As data scientists and R developers, we’re trained to think systematically, explore data thoroughly, and validate our assumptions. But here’s a reality check: your dashboard users don’t think like you do.

Consider this scenario: - You see: A well-organized dashboard with 15 filters, 8 visualizations, and comprehensive data coverage - Your users see: An overwhelming interface where they can’t find what they need quickly

This isn’t because your users are less intelligent; it’s because they’re human, and humans have predictable cognitive patterns that affect how they process information.

The Core Problem: Cognitive Load vs. Analytical Depth

What You’re Optimized For (as a data scientist)

# Your typical approach to data exploration
data |>
  filter(multiple_conditions) |>
  group_by(several_dimensions) |>
  summarize(
    metric_1 = mean(value_1, na.rm = TRUE),
    metric_2 = sum(value_2),
    metric_3 = median(value_3),
    .groups = "drop"
  ) |>
  arrange(desc(metric_1))

What Your Users Need (cognitive efficiency)

Quick orientation: “What am I looking at?”
Immediate value: “What does this mean for my decision?”
Clear next steps: “What should I do about it?”

Key Behavioral Science Concepts for Data People

1. Cognitive Load Theory: The Mental RAM Problem

Think of human attention like computer memory: limited and easily overloaded.

# Explore cognitive load concepts
cognitive_concepts <- bid_concepts("cognitive")

cognitive_concepts |>
  select(concept, description, implementation_tips) |>
  head(3)

For dashboards, this means:

Users can only process 5-9 pieces of information simultaneously
Every filter, chart, and button competes for mental resources
Progressive disclosure beats comprehensive display

Practical example:

# Instead of showing all 12 KPIs at once:
# kpi_grid <- layout_columns(
#   value_box("Revenue", "$1.2M", icon = "currency-dollar"),
#   value_box("Customers", "15,432", icon = "people"),
#   # ... 10 more value boxes
# )

# Show key metrics first, details on demand:
kpi_summary <- layout_columns(
  col_widths = c(8, 4),
  card(
    card_header("Key Performance Summary"),
    value_box("Primary Goal", "$1.2M Revenue", icon = "target"),
    p("vs. $980K target (+22%)")
  ),
  card(
    card_header("Details"),
    actionButton(
      "show_details",
      "View All Metrics",
      class = "btn-outline-primary"
    )
  )
)

2. Anchoring Bias: The First Number Problem

Humans over-rely on the first piece of information they see. In data dashboards, this means the first number shown becomes a reference point for everything else.

# Learn about anchoring
bid_concept("anchoring") |>
  select(description, implementation_tips)

Dashboard implication:

If you show “Sales: $50K” first, users will judge all subsequent numbers relative to $50K, even if it’s not the most important metric.

Solution pattern:

# Provide context and reference points
sales_card <- card(
  card_header("Monthly Sales Performance"),
  layout_columns(
    value_box(
      title = "This Month",
      value = "$87K",
      showcase = bs_icon("graph-up"),
      theme = "success"
    ),
    div(
      p("Previous month: $65K", style = "color: #666; margin: 0;"),
      p("Target: $80K", style = "color: #666; margin: 0;"),
      p(strong("vs. Target: +9%"), style = "color: #28a745;")
    )
  )
)

3. Framing Effects: How You Say It Matters More Than What You Say

The same data can be interpreted completely differently based on how it’s presented.

# Explore framing concepts
bid_concept("framing") |>
  select(description, implementation_tips)

Example: Customer satisfaction of 73%

# Negative frame (emphasizes problems)
satisfaction_negative <- value_box(
  "Customer Issues",
  "27% Unsatisfied",
  icon = "exclamation-triangle",
  theme = "danger"
)

# Positive frame (emphasizes success)
satisfaction_positive <- value_box(
  "Customer Satisfaction",
  "73% Satisfied",
  icon = "heart-fill",
  theme = "success"
)

# Balanced frame (shows progress)
satisfaction_balanced <- card(
  card_header("Customer Satisfaction Progress"),
  value_box("Current Level", "73%"),
  p("Improvement needed: 17 percentage points to reach 90% target")
)

4. Choice Overload: Why More Options Create Worse Outcomes

Research shows that too many choices lead to decision paralysis and user dissatisfaction.

The data scientist instinct:

“Let’s give them 20 different chart types and 15 filters so they can explore any question!”

The user reality:

Use defaults and ignore customization options
Get overwhelmed and abandon the task
Make suboptimal choices due to cognitive fatigue

Better approach:

# Instead of 15 filters visible at once
ui_complex <- div(
  selectInput("region", "Region", choices = regions),
  selectInput("product", "Product", choices = products),
  selectInput("channel", "Channel", choices = channels),
  # ... 12 more filters
)

# Use progressive disclosure
ui_simple <- div(
  # Show only the most common filters first
  selectInput("time_period", "Time Period", choices = time_options),
  selectInput("metric", "Primary Metric", choices = key_metrics),

  # Advanced filters behind a toggle
  accordion(
    accordion_panel(
      "Advanced Filters",
      icon = bs_icon("sliders"),
      selectInput("region", "Region", choices = regions),
      selectInput("product", "Product", choices = products)
      # Additional filters here
    )
  )
)

Translating Data Science Skills to UX Thinking

Your A/B Testing Mindset → UX Validation

You already know how to test hypotheses with data. Apply the same rigor to UX decisions:

# Your typical A/B test
results <- t.test(
  treatment_group$conversion_rate,
  control_group$conversion_rate
)

# UX equivalent: Test interface variations
dashboard_test <- list(
  control = "Current 5-chart overview page",
  treatment = "Redesigned with progressive disclosure"
)

# Measure: task completion time, user satisfaction, error rates
# Analyze: same statistical rigor you'd apply to any experiment

Your Data Validation Process → User Mental Model Validation

Just as you validate data quality, validate that your interface matches user expectations:

# Document user mental models like you document data assumptions
user_assumptions <- bid_interpret(
  central_question = "How do sales managers think about performance?",
  data_story = list(
    hook = "Sales managers need quick performance insights",
    context = "They have 15 minutes between meetings",
    tension = "Current reports take too long to interpret",
    resolution = "Provide immediate visual context with drill-down capability"
  )
)

# Validate these assumptions like you'd validate data quality
summary(user_assumptions)

The BID Framework: A Systematic Approach for Data People

The BID framework gives you a systematic way to apply behavioral science, similar to how you approach data analysis:

Stage 1: Interpret (Like defining your research question)

What specific question does this dashboard answer?
Who are the users and what are their constraints?

Stage 2: Notice (Like identifying data quality issues)

Where do users get stuck or confused?
What cognitive bottlenecks exist?

Stage 3: Anticipate (Like checking for statistical biases)

What cognitive biases might affect interpretation?
How can design mitigate these biases?

Stage 4: Structure (Like choosing the right visualization)

What layout best supports the user’s mental model?
How should information be hierarchically organized?

Stage 5: Validate (Like testing your model)

Does the user leave with clear insights?
Can they take action on what they’ve learned?

Practical Next Steps

Start small: Pick one dashboard and apply the BID framework
Measure: Track user behavior metrics (time to insight, task completion)
Iterate: Use the same experimental mindset you apply to model development
Learn progressively: Use bid_concepts() to explore one behavioral principle at a time

# Explore available concepts by category
all_concepts <- bid_concepts()
table(all_concepts$category)

# Start with these fundamental concepts for data dashboards
starter_concepts <- c(
  "Cognitive Load Theory",
  "Anchoring Effect",
  "Processing Fluency",
  "Progressive Disclosure"
)

for (concept in starter_concepts) {
  cat("\n### ", concept, "\n")
  info <- bid_concept(concept)
  cat(info$description[1], "\n")
  cat("Implementation:", info$implementation_tips[1], "\n")
}

Why This Matters for You

Understanding user cognition makes you a more complete data professional:

Better adoption: Your analyses get used instead of ignored
Clearer communication: Stakeholders understand and act on your insights
Strategic thinking: You solve business problems, not just technical problems
Competitive advantage: Most data scientists don’t think about user experience

Remember: The best analysis in the world is worthless if users can’t understand and act on it. Behavioral science helps bridge that gap.

Resources for Further Learning

Use bid_concepts() to explore all available behavioral science concepts
Try the getting-started vignette for hands-on BID framework practice
Check out the telemetry-integration vignette to measure the impact of your UX improvements

The goal isn’t to become a UX expert; it’s to apply your analytical thinking to user experience and create more effective data products.

These binaries (installable software) and packages are in development.
They may not be fully stable and should be used with caution. We make no claims about them.