Demo: Using create_odds_ratios from the vivainsights Python Package

This notebook demonstrates how to use the create_odds_ratios function from the vivainsights Python package to analyze the relationship between ordinal metrics and an independent variable.

In this walkthrough, you will:

  1. Load demo data (pq_data) from the package.

  2. Create an independent variable (UsageSegments_12w) using identify_usage_segments.

  3. Compute favorability scores for ordinal metrics with compute_fav.

  4. Calculate odds ratios for ordinal metrics using create_odds_ratios.

  5. Visualize the results for easier interpretation.

[1]:

# Import necessary libraries
import vivainsights as vi
import pandas as pd
import warnings

# Suppress warnings for cleaner output
warnings.filterwarnings("ignore")

Step 1: Load the demo data

First, load the sample Person Query dataset (pq_data) provided by vivainsights.

[2]:

# Load the demo data
pq_data = vi.load_pq_data()

# Display the first few rows of the dataset
pq_data.head()

[2]:
PersonId MetricDate Collaboration_hours Copilot_actions_taken_in_Teams Meeting_and_call_hours Internal_network_size Email_hours Channel_message_posts Conflicting_meeting_hours Large_and_long_meeting_hours ... Summarise_chat_actions_taken_using_Copilot_in_Teams Summarise_email_thread_actions_taken_using_Copilot_in_Outlook Summarise_meeting_actions_taken_using_Copilot_in_Teams Summarise_presentation_actions_taken_using_Copilot_in_PowerPoint Summarise_Word_document_actions_taken_using_Copilot_in_Word FunctionType SupervisorIndicator Level Organization LevelDesignation
0 bf361ad4-fc29-432f-95f3-837e689f4ac4 2024-03-31 17.452987 4 11.767599 92 7.523189 0.753451 2.079210 0.635489 ... 2 0 0 0 0 Specialist Manager Level3 IT Senior IC
1 0500f22c-2910-4154-b6e2-66864898d848 2024-03-31 32.860820 6 26.743370 193 11.578396 0.000000 8.106997 1.402567 ... 2 0 4 1 0 Specialist Manager Level2 Legal Senior Manager
2 bb495ec9-8577-468a-8b48-e32677442f51 2024-03-31 21.502359 8 13.982031 113 9.073214 0.894786 3.001401 0.000192 ... 1 1 0 0 0 Manager Manager Level4 Legal Junior IC
3 f6d58aaf-a2b2-42ab-868f-d7ac2e99788d 2024-03-31 25.416502 4 16.895513 131 10.281204 0.528731 1.846423 1.441596 ... 0 0 0 0 0 Manager Manager Level1 HR Executive
4 c81cb49a-aa27-4cfc-8211-4087b733a3c6 2024-03-31 11.433377 4 6.957468 75 5.510535 2.288934 0.474048 0.269996 ... 0 0 1 0 0 Technician Manager Level1 Finance Executive

5 rows × 73 columns

Step 2: Create the independent variable with identify_usage_segments

Use identify_usage_segments to classify users into usage segments based on their Copilot actions. The independent variable (UsageSegments_12w) is created by aggregating columns that start with Copilot_actions_taken_in_.

[3]:

# Identify usage segments
usage_segments_data = vi.identify_usage_segments(
    data=pq_data,
    metric_str=[
        "Copilot_actions_taken_in_Teams",
        "Copilot_actions_taken_in_Outlook",
        "Copilot_actions_taken_in_Excel",
        "Copilot_actions_taken_in_Word",
        "Copilot_actions_taken_in_Powerpoint"
    ],
    version="12w",
    return_type="data"
)

# Display the first few rows of the updated dataset
usage_segments_data.head()

[3]:
PersonId MetricDate Collaboration_hours Copilot_actions_taken_in_Teams Meeting_and_call_hours Internal_network_size Email_hours Channel_message_posts Conflicting_meeting_hours Large_and_long_meeting_hours ... Level Organization LevelDesignation target_metric target_metric_l12w target_metric_l4w IsHabit12w IsHabit4w UsageSegments_12w UsageSegments_4w
0 01986072-719a-404c-ae98-009d92e82323 2024-03-31 26.884733 7 17.700027 156 9.667004 0.117751 2.674868 1.262361 ... Level4 IT Junior IC 10 10.00 10.00 False False Novice User Novice User
1 01986072-719a-404c-ae98-009d92e82323 2024-04-07 21.280727 10 15.372990 121 8.417014 0.519473 0.368913 2.108141 ... Level4 IT Junior IC 12 11.00 11.00 False False Novice User Novice User
2 01986072-719a-404c-ae98-009d92e82323 2024-04-14 17.450330 8 11.808617 104 7.889519 1.907069 0.096829 0.853150 ... Level4 IT Junior IC 11 11.00 11.00 False False Novice User Novice User
3 01986072-719a-404c-ae98-009d92e82323 2024-04-21 21.368059 3 14.908550 115 6.776404 0.209775 3.953832 0.878616 ... Level4 IT Junior IC 4 9.25 9.25 False True Novice User Habitual User
4 01986072-719a-404c-ae98-009d92e82323 2024-04-28 20.849744 5 13.737000 110 8.759793 0.931585 1.201305 0.000000 ... Level4 IT Junior IC 6 8.60 8.25 False True Novice User Habitual User

5 rows × 80 columns

Visualize the mean of target_metric by usage segment

To better understand usage behavior, create a bar plot showing the mean of target_metric grouped by UsageSegments_12w.

[4]:

# Visualize the mean of `target_metric` by `UsageSegments_12w`
usage_segments_bar_plot = vi.create_bar(
    data=usage_segments_data,
    metric="target_metric",
    hrvar="UsageSegments_12w",
    return_type="plot",
    plot_title="Mean Target Metric by Usage Segment",
    plot_subtitle="Based on 12-week rolling averages"
)

# Display the bar plot
usage_segments_bar_plot.show()
_images/demo-create_odds_ratios_7_0.png

Visualize usage segments over time

Next, visualize the distribution of usage segments over time using identify_usage_segments with return_type='plot'. The following shows a horizontal stacked bar plot, which shows the evolution in the proportion of the usage segments over time.

[8]:

# Visualize usage segments over time
usage_segments_time_plot = vi.identify_usage_segments(
    data=pq_data,
    metric_str=[
        "Copilot_actions_taken_in_Teams",
        "Copilot_actions_taken_in_Outlook",
        "Copilot_actions_taken_in_Excel",
        "Copilot_actions_taken_in_Word",
        "Copilot_actions_taken_in_Powerpoint"
    ],
    version="12w",
    return_type="plot"
)

# Display the time plot
usage_segments_time_plot.show()
_images/demo-create_odds_ratios_9_0.png

Step 3: Compute favorability scores for ordinal metrics

Before calculating odds ratios, use compute_fav() to convert ordinal metrics into categorical variables representing favorable and unfavorable scores. This standardizes metrics to a 100-point scale, making results easier to interpret and compare.

Neutral scores are dropped to focus on the most meaningful responses.

In usage_segments_data printed below, it can be seen that compute_fav() has added several columns suffixing the ordinal_metrics columns with _100 and _fav.

[9]:

# Define the ordinal metrics
ordinal_metrics = [
    "eSat",
    "Initiative",
    "Manager_Recommend",
    "Resources",
    "Speak_My_Mind",
    "Wellbeing",
    "Work_Life_Balance",
    "Workload"
]

# Compute favorability scores
usage_segments_data = vi.compute_fav(
    data=usage_segments_data,
    ord_metrics=ordinal_metrics,
    item_options=5,  # Assuming a 5-point scale for ordinal metrics
    fav_threshold=70,
    unfav_threshold=40,
    drop_neutral=True
)

# Display the first few rows of the updated dataset
usage_segments_data.head()

[9]:
PersonId MetricDate Collaboration_hours Copilot_actions_taken_in_Teams Meeting_and_call_hours Internal_network_size Email_hours Channel_message_posts Conflicting_meeting_hours Large_and_long_meeting_hours ... Resources_100 Resources_fav Speak_My_Mind_100 Speak_My_Mind_fav Wellbeing_100 Wellbeing_fav Work_Life_Balance_100 Work_Life_Balance_fav Workload_100 Workload_fav
36 02723512-4f45-4385-8d1a-c23048e1e961 2024-04-07 26.310260 1 17.635230 124 10.887553 0.000000 3.322255 0.067661 ... 25.0 unfav 25.0 unfav 100.0 fav 0.0 unfav 0.0 unfav
83 02c55079-f137-4abb-9806-f58e9b60efd6 2024-06-30 17.401642 4 10.399207 84 5.253439 0.195852 3.203440 0.975272 ... 25.0 unfav 25.0 unfav 100.0 fav 0.0 unfav 0.0 unfav
123 02ddc980-8f37-4156-9397-6d621e445a00 2024-08-04 20.612899 3 14.130869 103 8.070390 0.577123 1.374351 0.000000 ... 25.0 unfav 25.0 unfav 100.0 fav 0.0 unfav 0.0 unfav
135 02ddc980-8f37-4156-9397-6d621e445a00 2024-10-27 19.514361 2 10.986860 91 6.221707 2.286118 2.294472 0.391576 ... 25.0 unfav 25.0 unfav 100.0 fav 0.0 unfav 0.0 unfav
164 032432ad-390c-4ce4-9f25-d5be080bd982 2024-09-15 34.160594 3 27.364673 182 12.926987 0.197464 6.306590 1.153810 ... 25.0 unfav 25.0 unfav 100.0 fav 0.0 unfav 0.0 unfav

5 rows × 96 columns

Step 4: Calculate odds ratios for ordinal metrics

Now, calculate odds ratios for the favorability scores of these ordinal metrics:

  • eSat

  • Initiative

  • Manager_Recommend

  • Resources

  • Speak_My_Mind

  • Wellbeing

  • Work_Life_Balance

  • Workload

The independent variable is UsageSegments_12w.

[13]:

# Calculate odds ratios
odds_ratios_table = vi.create_odds_ratios(
    data=usage_segments_data,
    ord_metrics=ordinal_metrics,
    metric="UsageSegments_12w",
    return_type="table"
)

# Display the odds ratios table
print(odds_ratios_table)

   UsageSegments_12w Level  Odds_Ratio     Ordinal_Metric      n
0      Habitual User     1    1.000000               eSat    3.0
1        Novice User     1    1.000000               eSat    1.0
2      Habitual User     2   53.571429               eSat  135.0
3        Novice User     2   37.000000               eSat   49.0
4      Habitual User     4   18.142857               eSat   58.0
5        Novice User     4   10.333333               eSat   14.0
6      Habitual User     5    0.428571               eSat    1.0
7        Novice User     5    0.333333               eSat    NaN
8      Habitual User     1    1.000000         Initiative    4.0
9        Novice User     1    1.000000         Initiative    3.0
10     Habitual User     2   54.333333         Initiative  166.0
11       Novice User     2   18.142857         Initiative   57.0
12     Habitual User     4    1.444444         Initiative    6.0
13       Novice User     4    1.571429         Initiative    5.0
14     Habitual User     1    1.000000  Manager_Recommend    5.0
15       Novice User     1    1.000000  Manager_Recommend    NaN
16     Habitual User     2   43.545455  Manager_Recommend  163.0
17       Novice User     2  133.000000  Manager_Recommend   59.0
18     Habitual User     4    1.909091  Manager_Recommend   10.0
19       Novice User     4    7.000000  Manager_Recommend    3.0
20     Habitual User     5    0.090909  Manager_Recommend    NaN
21       Novice User     5    5.000000  Manager_Recommend    2.0
22     Habitual User     1    1.000000          Resources    2.0
23       Novice User     1    1.000000          Resources    NaN
24     Habitual User     2  100.600000          Resources  171.0
25       Novice User     2  141.000000          Resources   62.0
26     Habitual User     4    0.600000          Resources    1.0
27       Novice User     4    3.000000          Resources    1.0
28     Habitual User     1    1.000000      Speak_My_Mind    2.0
29       Novice User     1    1.000000      Speak_My_Mind    1.0
30     Habitual User     2   97.400000      Speak_My_Mind  166.0
31       Novice User     2   43.666667      Speak_My_Mind   58.0
32     Habitual User     4    3.800000      Speak_My_Mind    9.0
33       Novice User     4    3.666667      Speak_My_Mind    5.0
34     Habitual User     4    1.000000          Wellbeing   74.0
35       Novice User     4    1.000000          Wellbeing   24.0
36     Habitual User     5    1.786885          Wellbeing  123.0
37       Novice User     5    1.938776          Wellbeing   43.0
38     Habitual User     1    1.000000  Work_Life_Balance  143.0
39       Novice User     1    1.000000  Work_Life_Balance   49.0
40     Habitual User     2    0.352785  Work_Life_Balance   58.0
41       Novice User     2    0.321101  Work_Life_Balance   17.0
42     Habitual User     1    1.000000           Workload  143.0
43       Novice User     1    1.000000           Workload   51.0
44     Habitual User     2    0.317829           Workload   54.0
45       Novice User     2    0.252174           Workload   14.0

Since favorability columns with the values fav, unfav, and neu have already been created using compute_fav(), you can use these directly in the proportional odds model to simplify the analysis.

When interpreting odds ratios, a value greater than 1 indicates that the odds of a favorable outcome are higher for the group compared to the reference group, while a value less than 1 means the odds are lower. An odds ratio of exactly 1 suggests no difference between groups. This helps you understand how different usage segments are associated with the likelihood of favorable responses on each metric.

[14]:

# Define ordinal metrics with '_fav' suffix
ordinal_metrics_fav = [f"{metric}_fav" for metric in ordinal_metrics]

# Calculate odds ratios
odds_ratios_table_fav = vi.create_odds_ratios(
    data=usage_segments_data,
    ord_metrics=ordinal_metrics_fav,
    metric="UsageSegments_12w",
    return_type="table"
)

# Display the odds ratios table
print(odds_ratios_table_fav)

   UsageSegments_12w  Level  Odds_Ratio         Ordinal_Metric    n
0      Habitual User    fav    1.000000               eSat_fav   59
1        Novice User    fav    1.000000               eSat_fav   14
2      Habitual User  unfav    2.953488               eSat_fav  137
3        Novice User  unfav    3.645161               eSat_fav   50
4      Habitual User    fav    1.000000         Initiative_fav    6
5        Novice User    fav    1.000000         Initiative_fav    5
6      Habitual User  unfav   38.230769         Initiative_fav  168
7        Novice User  unfav   12.090909         Initiative_fav   59
8      Habitual User    fav    1.000000  Manager_Recommend_fav   10
9        Novice User    fav    1.000000  Manager_Recommend_fav    5
10     Habitual User  unfav   23.285714  Manager_Recommend_fav  167
11       Novice User  unfav   12.090909  Manager_Recommend_fav   59
12     Habitual User    fav    1.000000          Resources_fav    1
13       Novice User    fav    1.000000          Resources_fav    1
14     Habitual User  unfav  169.000000          Resources_fav  172
15       Novice User  unfav   47.000000          Resources_fav   62
16     Habitual User    fav    1.000000      Speak_My_Mind_fav    9
17       Novice User    fav    1.000000      Speak_My_Mind_fav    5
18     Habitual User  unfav   25.842105      Speak_My_Mind_fav  167
19       Novice User  unfav   12.090909      Speak_My_Mind_fav   59
20     Habitual User    fav    1.000000          Wellbeing_fav  172
21       Novice User    fav    1.000000          Wellbeing_fav   63
22     Habitual User  unfav    1.000000  Work_Life_Balance_fav  172
23       Novice User  unfav    1.000000  Work_Life_Balance_fav   63
24     Habitual User  unfav    1.000000           Workload_fav  172
25       Novice User  unfav    1.000000           Workload_fav   63

[ ]:

# Filter for Level == 'fav' only, and sort Odds_Ratio in descending order
odds_ratios_table_fav = odds_ratios_table_fav[odds_ratios_table_fav['Level'] == 'fav']
odds_ratios_table_fav = odds_ratios_table_fav.sort_values(by='Odds_Ratio', ascending=False)

print(odds_ratios_table_fav)

   UsageSegments_12w Level  Odds_Ratio         Ordinal_Metric    n
4      Habitual User   fav         1.0         Initiative_fav    6
5        Novice User   fav         1.0         Initiative_fav    5
8      Habitual User   fav         1.0  Manager_Recommend_fav   10
16     Habitual User   fav         1.0      Speak_My_Mind_fav    9
9        Novice User   fav         1.0  Manager_Recommend_fav    5
12     Habitual User   fav         1.0          Resources_fav    1
1        Novice User   fav         1.0               eSat_fav   14
17       Novice User   fav         1.0      Speak_My_Mind_fav    5
13       Novice User   fav         1.0          Resources_fav    1
21       Novice User   fav         1.0          Wellbeing_fav   63
0      Habitual User   fav         1.0               eSat_fav   59
20     Habitual User   fav         1.0          Wellbeing_fav  172

Step 5: Visualize the odds ratios

Create a bar plot to visualize the odds ratios for the ordinal metrics, making it easier to compare the impact of usage segments.

[20]:

# Visualize odds ratios
odds_ratios_plot = vi.create_odds_ratios(
    data=usage_segments_data,
    ord_metrics=ordinal_metrics,
    metric="UsageSegments_12w",
    return_type="plot"
)

# Display the plot
odds_ratios_plot.show()
_images/demo-create_odds_ratios_18_0.png 
[22]:

# Visualize odds ratios for favorability
odds_ratios_plot_fav = vi.create_odds_ratios(
    data=usage_segments_data,
    ord_metrics=ordinal_metrics_fav,
    metric="UsageSegments_12w",
    return_type="plot"
)

# Display the plot
odds_ratios_plot_fav.show()
_images/demo-create_odds_ratios_19_0.png

Summary

In this notebook, you learned how to:

  1. Load demo data (pq_data).

  2. Create an independent variable (UsageSegments_12w) using identify_usage_segments.

  3. Compute favorability scores for ordinal metrics with compute_fav.

  4. Calculate odds ratios for ordinal metrics using create_odds_ratios.

  5. Visualize the results for interpretation.

By combining create_odds_ratios with compute_fav, you can consistently analyze the relationship between ordinal metrics and independent variables, regardless of the original point scale.