Kategori: Data Literacy

There is one chart that almost inevitably makes the rounds every time Statistics Indonesia (BPS) releases labor force figures: the bar chart showing unemployment rates by educational attainment. It often circulates with provocative headlines like “University Graduates Face Higher Unemployment Than Primary School Leavers” or “Higher Education Offers No Guarantee of Employment.”

The numbers are accurate. The conclusions, however, are almost always flawed. Today, we unpack why—and explore how to present group comparison data with integrity.

The Data

Here is the Open Unemployment Rate (TPT) for February 2026 based on the highest level of education completed, drawn from the BPS National Labor Force Survey (Sakernas):

Highest education completed	TPT February 2026	TPT May 2026
Vocational High School (SMK)	7.74%	7.68%
Senior High School (SMA)	6.23%	6.17%
Diploma IV, Bachelor’s, Master’s, Doctorate	6.13%	6.09%
Diploma I/II/III	4.80%	4.75%
Junior High School (SMP)	4.19%	4.16%
Primary School or below	2.32%	2.31%

The pattern holds steady across both periods: vocational graduates (SMK) register the highest rate, those with primary education or below show the lowest, and university graduates sit noticeably above junior high leavers.

Turn these figures into a standard bar chart sorted from highest to lowest, and the visual impression is stark: the less educated you are, the lower your unemployment rate appears. That intuitive takeaway is precisely what demands a closer look.

Trap 1: The unemployment rate does not measure job quality

Remember the official survey definition: a person is classified as employed if they worked for at least one hour during the previous week to earn an income. This threshold is intentionally broad.

Individuals with lower educational attainment often cannot afford the luxury of waiting for an ideal job. Immediate living expenses mean they must take whatever work is readily available: casual day labor, unpaid family work, small-scale petty trade, or other informal gigs offering meager pay and no social safety net. In official statistics, they are employed. In reality, many survive under highly precarious conditions.

Conversely, university graduates are far more likely to receive temporary financial backing from family, allowing them to hold out for roles that match their degree and qualifications. While searching, they are counted as unemployed.

A lower unemployment rate among less-educated groups primarily reflects an inability to remain unemployed, rather than genuine success in securing decent work.

Trap 2: A percentage is not a headcount

A 7.74% unemployment rate for vocational high school graduates is calculated solely from the labor force within that specific educational cohort. It answers one question: “Among vocational graduates who are employed or actively seeking work, what percentage is unemployed?”

It does not answer: “Among all unemployed individuals nationwide, which educational group is the largest?” For that question, you need absolute counts or the overall distribution of unemployed individuals by education.

Both questions are legitimate and valuable, but they require entirely different metrics. The most common error in public discourse is answering a composition question with a cohort rate.

Trap 3: Ignoring age differences across cohorts

University graduates typically enter the formal labor market later in life, usually around ages 22 to 23. By contrast, the labor force holding only primary education is heavily concentrated in older age brackets—workers who have spent decades in the workforce. As labor data consistently shows, youth unemployment is naturally much higher than adult unemployment across the board.

Consequently, part of the observed difference between educational tiers is driven by age demographics, not the educational level itself. A fairer comparison evaluates educational cohorts within the same age bracket.

How to present this chart with honesty

For data analysts, journalists, and researchers tasked with communicating group comparisons, several straightforward principles help preserve nuance:

1. State the denominator explicitly in the title. Avoid ambiguous headlines like “Unemployment by Education.” Opt instead for “Open Unemployment Rate (Percent of Labor Force per Educational Group), February 2026.” A slightly longer title is far preferable to a misleading one.

1. Display rates and absolute headcounts side by side. Two compact visual panels—one showing percentages and the other total headcounts—instantly resolve two fundamentally distinct questions.

1. Include a brief note on definitions. A single line beneath the chart—”Employed: worked at least one hour during the past week for income”—contextualizes why lower-educated groups appear so well-positioned.

1. Add quality-of-work indicators when possible. Incorporating the share of informal employment or average earnings per educational tier provides essential context. Without it, the chart delivers only half the story.

1. Start your axes at zero. Truncating the baseline on bar charts exaggerates minor variances. A shift between 6.13% and 6.23% looks dramatic if the axis starts at 6.0%.

1. Exercise caution with sorting. Sorting bars strictly from highest to lowest facilitates ranking, but also implies hierarchy. When educational tiers possess a natural sequence, preserving that progressive order often makes non-linear patterns immediately visible.

Lessons for organizational reporting

The identical phenomenon surfaces across corporate and departmental dashboards. A chart showing “complaint rates per branch” can make smaller branch offices look disproportionately troublesome due to small denominators. A chart showing “average sales per representative” can unfairly penalize junior representatives who joined just three months ago. A metric of “absence rates across business units” can deceive if units classify leave differently.

Before any group comparison slide makes its way into executive meetings, ensure it passes three essential questions:

• What is the exact denominator, and is it stated clearly?
• Are confounding variables at play across groups, such as age, tenure, or headcount?
• Does the visualization directly address the strategic question decision-makers are actually asking?

A quick exercise for data teams

If you lead an analytics or BI team, try this brief exercise in your next session. Take an existing group comparison chart from a recent report and ask each team member to write down one single-sentence takeaway without consulting one another. Then compare their responses.

Almost invariably, the conclusions will diverge. Someone will mistake a cohort rate for a national total, overlook the denominator, or draw unwarranted causal inferences from a simple descriptive comparison. That variance does not mean your team lacks skill; it simply means the visual design has room for improvement.

Use those insights to refine the graphic: add the denominator to the title, clarify definitions, and introduce headcount totals where helpful. Repeat the exercise, and you will notice far more consistent takeaways. A well-designed chart leads different people to the same valid conclusion.

Closing thoughts

The chart of unemployment by education is a prime example of how accurate figures can generate misleading conclusions. It is rarely a matter of deliberate deception; rather, we burden a simple metric with questions it was never designed to answer.

A higher unemployment rate among vocational graduates remains an issue warranting policy attention, as vocational curricula are explicitly structured for direct workforce readiness. But low unemployment among primary school leavers is hardly a cause for celebration. Both figures must be interpreted alongside measures of employment quality.

An honest chart may not always be the most sensational, but it is always the most dependable.

Read also: looking to disaggregate and visualize your organization’s survey or operational data? Explore our Data Analysis & Statistical Services.

---

Every time the Central Statistics Agency (BPS) publishes labor force data, one number immediately floods chat groups and social media timelines. For February 2026, that number was the open unemployment rate (TPT) for the 15–24 age group: 16.36%. Compare that to the national headline rate of just 4.68%. Three months later, in May 2026, the youth figure climbed again to 17.37%, while the national average nudged down slightly to 4.65%.

The sentence most often attached to that statistic: “16% of Indonesia’s youth are unemployed.” It sounds reasonable at first glance, yet it is fundamentally misleading. And small misunderstandings like this can easily lead to misguided public policies.

This article unpacks how to read the open unemployment rate properly: what it actually defines, what sits in its denominator, and the most common pitfalls when comparing it across groups.

The formula is simple—the denominator is what people forget

The formula for the unemployment rate is straightforward:

Unemployment Rate = Number of Unemployed ÷ Total Labor Force × 100%

What people often overlook is the denominator: the labor force. The labor force includes working-age individuals who are currently employed plus those actively looking for work or preparing to start a business. People who are attending school, studying at university, managing a household, or not seeking employment for other reasons are excluded from the labor force altogether.

In the 15–24 age bracket, the majority of young people are still in school or college. As a result, the labor force denominator for this group is substantially smaller than the total youth population.

So the accurate statement is not “16% of youth are unemployed,” but rather: among young people aged 15–24 who are working or actively seeking work, around 16 out of 100 have not yet found a job.

The difference may seem subtle, but the distinction matters. The first statement creates the impression of an overwhelming generational crisis. The second points to a much more specific challenge: a narrow gateway into the workforce for those who are already ready and looking to work.

The definition of “employed” that often surprises people

Official employment figures come from the National Labor Force Survey (Sakernas). BPS surveyors visit sampled households and record the primary activities of every household member during the previous week.

Under the survey’s official definition, a person is classified as employed if they worked for at least one hour during the past week to earn income. This one-hour threshold adheres to international standards (ILO) and is intentionally broad to capture informal and intermittent economic activity.

The consequence: someone doing odd jobs for a few hours a week, assisting in a family stall, or working casual part-time gigs for very modest pay is still categorized as employed. They are not counted among the unemployed.

This is precisely why a low headline unemployment rate does not automatically signal healthy labor market conditions. To assess job quality, we need complementary indicators such as hours worked, underemployment, formal versus informal status, and wage levels.

Summary table: numbers that should always be read together

Here are the key open unemployment figures across age groups from Sakernas, as published by BPS:

Group	November 2025	February 2026	May 2026
National	4.74%	4.68%	4.65%
Age 15–24	16.26%	16.36%	17.37%
Age 25–59	2.98%	2.93%	2.56%
Age 60 and above	1.44%	1.89%	2.35%

The pattern is striking: while the national rate remained steady with a slight downward trend, the youth rate consistently moved higher. Looking only at the national aggregate causes decision-makers to miss the most critical story.

Four common traps when interpreting unemployment data

Trap 1: Confusing the unemployment rate with the share of the entire population. As discussed earlier, always ask: a percentage of what? The denominator dictates the meaning.

Trap 2: Comparing different survey periods without accounting for seasonality. Sakernas is conducted multiple times a year. Employment patterns have distinct seasonal rhythms: harvest seasons, holiday periods, and school graduation cycles. Comparing February directly with August without noting seasonal factors can easily mislead. The soundest approach is comparing the same period year-over-year, or explicitly stating the months compared.

Trap 3: Overlooking subgroup sample size. Large-scale surveys represent broad populations reliably, but as subgroups narrow, the margin of error widens. The youth unemployment rate in a single regency carries much greater uncertainty than the national benchmark. Small shifts in small sub-samples may simply be statistical noise.

Trap 4: Inferring causality from a single isolated metric. High youth unemployment can stem from many competing drivers: the transition gap from school to work, skills mismatches, asymmetric job vacancy information, strict experience requirements for entry-level roles, or overall macroeconomic sluggishness. Each hypothesis requires different empirical evidence to prove. A single rate cannot tell you which mechanism is at play.

Why youth numbers are almost always higher

Across almost every country, youth unemployment rates sit consistently above national averages. This is not unique to Indonesia. Standard labor economics literature highlights several structural reasons:

1. School-to-work transition. New graduates lack workplace experience and professional networks. Securing a first role inherently takes time.
2. Search and matching. Many young jobseekers still have family support, enabling them to wait for an opportunity aligned with their education. Statistically, this search period is classified as unemployment.
3. Last in, first out. When firms scale back hiring or trim headcount, employees with shorter tenure are typically impacted first.

Because the structural gap is common, what matters most is not simply whether youth rates exceed the average, but their trajectory. And through the first half of 2026, the trend has been upward.

Practical takeaways for decision-makers and teams

The principles of proper denominators and subgroup breakdowns apply far beyond macro labor statistics. Inside organizations, the exact same dynamics occur daily.

A company-wide turnover rate of 12% may look reassuring, yet turnover among first-year hires might be alarming. An overall customer satisfaction score of 85% might seem solid, while a critical enterprise client segment is deeply dissatisfied. Averages soothe; disaggregated data reveals where the friction actually lives.

Three practical habits I recommend:

• State the denominator clearly in chart and table titles. “Percentage of labor force in each age bracket” is far more informative than simply “percent.”
• Disaggregate before concluding. Break down data by age cohort, region, tenure, or tier before making strategic calls.
• Pair headline metrics with quality indicators. Review unemployment alongside hours worked, job stability, and wage distributions.

Closing thoughts

The figure of 16.36% is concerning not merely because of its size, but because of its trajectory and meaning. It indicates that the gateway into formal employment for young people ready to work remains tight—and three months later, grew tighter still.

Reading that number accurately is the essential first step toward meaningful solutions. Not moral lectures on work ethic, but concrete improvements in the pathways linking education to first-time employment.

Read also: need support analyzing or interpreting your survey and business data? Explore our Data Analysis & Statistical Services.

---

A few weeks ago I wrote that only about one in five crime victims reports to the police. Since then, many people have asked the same question: “Okay, I understand the number. So what can be done?”

It’s a good question, and frankly, it’s harder to answer than understanding the number itself. This Sunday, to close the week, I’ll try to answer it in three layers: what we need to understand, who can do what, and what we should avoid.

First, recall the number

According to the 2024 National Socioeconomic Survey (Susenas), 20.28% of crime victims reported to the police. A year earlier it was 23.36%. That means roughly four out of five victims chose not to report, and the share who do report has actually fallen.

The impact on how we read statistics is significant. The police data we see in the news is only the part that makes it into the records. As a very rough illustration: if one type of crime is recorded at 1,000 cases and only 20% of its victims report, the actual number of victims could be around 5,000. This is not a finding, only a sense of scale, because reporting rates differ across crime types and regions.

Second, why don’t people report?

Before discussing action, we need to understand the causes. In the criminology literature, the most common reasons include:

• Seeing the loss as small, or the matter as too trivial to report.
• Not believing a report will be followed up, or worrying the process will be complicated.
• Fear or shame, especially for sensitive incidents or ones involving people close to them.
• Cost and time: the distance to the police station, the queue, and having to come back for further questioning.

Interestingly, when we put this data side by side with infrastructure data, the picture becomes even more unusual. Village Potential (Podes) data shows that 89.14% of villages and urban wards have no police post, yet around 85% of them say access to the police is easy. If distance isn’t the main barrier, then what needs closer study is trust and the experience of reporting.

Third, who can do what

Party	Reasonable actions
Statistical agencies and researchers	Keep publishing victimization surveys and link them to registration data
Law enforcement and complaint services	Make reporting easier and faster, keep reporters updated on their cases, train officers to be welcoming to people who report
Local governments	Provide alternative reporting channels (online, through community services) and monitor reporting rates, not just the number of reports
Media	Include the dark-figure context when reporting rises or falls in crime figures
Readers	Weigh crime figures against comparison sources before sharing them

A few brief explanations.

For researchers and statistical agencies, the key is triangulation. One source is never enough. Victimization surveys capture experience, registration data captures the system’s response, and perception surveys capture the sense of safety. Putting all three into one framework gives a far more honest picture.

For law enforcement and policymakers, there’s one trap to watch for. If a program to build public trust succeeds, the number of reports may rise at first, because people who used to stay silent start to come forward. If success is measured only by a drop in reports, a program that is actually working may look like a failure. Monitor the reporting rate, not just the number of reports.

For the media and readers, the most realistic action is a simple habit: every time a story says crime has “gone up” or “gone down,” ask whether what changed is the incidents themselves or people’s willingness to report them.

Fourth, what to avoid

There are three common mistakes when people talk about the dark figure of crime.

Blaming victims who don’t report. The decision not to report usually makes sense from the perspective of the person who experienced it. Calling it “not caring” or “not law-abiding” doesn’t help; it only widens the gap between victims and the system.

Turning an illustrative number into a fact. The “1,000 divided by 0.20 equals 5,000” calculation is deliberately simplified. Don’t quote it as if it were an official estimate.

Concluding that crime figures are useless. Registration data remains important, especially for understanding the workload of law enforcement and for comparisons over time with consistent definitions. What we need is to read it correctly, not to throw it away.

Applying it in organizations: from the dark figure to a reporting culture

The concept of a dark figure doesn’t belong to criminology alone. Imagine a company that records two workplace incident reports in a month. Management is pleased: “Great, our office is safe.” But there are two very different possibilities behind that number. Perhaps only two incidents really happened. Or perhaps employees are reluctant to report because the procedure is complicated, they’re afraid of being labeled troublemakers, or they aren’t sure their report will be used.

The way to tell them apart is similar to how BPS complements police data with surveys: run an anonymous survey asking whether employees have experienced or witnessed an incident in the past three months and whether they reported it. Compare the results with the number of official reports. The gap between the two is your organization’s dark figure. And that number, just like in crime statistics, is the most valuable information of all.

Good policy doesn’t stop at “reduce the number of incidents.” It first asks, “do people feel safe reporting?”, because without that, incident counts can’t be trusted as a measure of success.

Research questions worth exploring

For students and researchers looking for a topic, the dark figure opens many directions. What factors are associated with the decision to report, by type of crime? Is there a difference in reporting rates between urban and rural areas, and is the gap related to perceptions of law enforcement? How does the reporting rate change after a new complaint service is introduced? These questions can be answered with available survey data, with care for their limitations and research ethics.

Lessons for data readers

If we step outside criminology, this pattern shows up everywhere. A small number of customer complaints doesn’t necessarily mean customers are satisfied; maybe the complaint channel is hard to use. A small number of workplace incident reports doesn’t necessarily mean the office is safe; maybe people are afraid of being labeled troublemakers. “Clean” data sometimes means no one dares to record anything.

That’s why my favorite question whenever I see a small number is: who actually had the chance to show up in this data?

Closing

The dark figure can’t be eliminated, but it can be reduced and understood. The first step is to acknowledge that official records are not the whole of reality. The next step is to build systems that make people feel safe and confident enough to report, and to measure them with the right indicators.

Tomorrow we start a new week on young people and the world of work. Thank you for following this week’s discussions. If there’s a criminology or social data topic you’d like us to cover, leave it in the comments.

Sunday morning is a good time to pause and ask: what did we actually learn this week? Since Monday, 19 October, we’ve unpacked numbers from seven different angles: summarizing variables, environmental crime, choosing a statistical test, police posts, prediction models, crimes against life, questionnaires, and a checklist for reading crime figures. In the last two days, we also opened a new series on common misunderstandings of official statistics, plus a Q&A session.

If we squeeze it all down, there are seven lessons that I think are most worth taking home. Not the formulas, but the thinking habits.

1. More variables doesn’t mean more information

On Monday we looked at twenty-five variables that actually overlapped with each other. Factor analysis and PCA helped condense them into a few dimensions without losing the story. The lesson: before adding variables, ask whether they really measure different things. Summarizing well is often more useful than collecting more.

2. Enforcement figures measure system activity

Environmental crime was recorded at 4,149 incidents in 2024, and about 32% were concentrated in a single province. Concluding that this province is “the most damaged” is too quick. Crimes of this kind are only recorded when officers find them and act on them, so the figure also reflects how active enforcement is there. The lesson: distinguish between what happens and what gets found.

3. Method follows the question, not the other way around

On Tuesday, five questions guided us in choosing a statistical test: what the analysis is for, what the measurement scale is, how many groups there are and whether they’re paired, whether the assumptions hold, and how large the sample is. Students who pick a method first and then look for a question almost always end up rewriting their results chapter.

4. Access isn’t the only barrier

About 89.14% of villages and urban wards have no police post, yet around 85% say access is easy, and only about 20% of victims report. If physical access isn’t the main barrier, perhaps what needs building is trust. The lesson: don’t confuse the availability of a facility with its use.

5. High accuracy can mislead

A model predicting employee resignations with 85% accuracy sounds impressive, until we realize that if only 15% of employees resign, simply guessing “no one resigns” is already 85% accurate, with 0% recall. The lesson: always ask for complementary measures (precision and recall) and compare against a simple baseline.

6. Big numbers and small numbers both need context

Murder cases were recorded at 1,106 in 2024, down from 1,129. With numbers this small, minor fluctuations look large in percentage terms. Meanwhile, incidents of negligence causing death fell from 875 to 563. The lesson: for categories with small counts, be careful about drawing trends from a single year.

7. Data quality starts with the question

On Thursday we looked at seven questionnaire traps: double-barreled questions, leading questions, vague terms, absolute words, overlapping answer options, order effects, and questionnaires that are too long. No analysis, however sophisticated, can rescue a bad questionnaire. The lesson: the quality of an analysis is decided upstream, when the questions are written.

This week at a glance

Day	Topic	Lesson
Monday	Factor analysis and PCA	Condense overlapping variables
Monday	Environmental crime	Enforcement statistics measure system activity
Tuesday	Choosing a statistical test	Five questions before choosing a tool
Tuesday	Police posts and reporting	Access isn’t the only barrier
Wednesday	Resignation prediction model	Accuracy alone isn’t enough
Wednesday	Crimes against life	Small numbers need extra care
Thursday	Questionnaires and checklist	Data quality starts with the question
Friday and Saturday	Official statistics and Q&A	The numbers are right; the reading is often wrong

How to use these seven lessons at work

Lessons are only useful when they turn into habits. Here are a few simple ways to start next week.

In meetings: before accepting a number from a presentation, ask one clarifying question, such as “what is this calculated from?” or “what’s the comparison?”. Don’t make it rhetorical; ask out of curiosity, and presenters are usually glad, because it shows their numbers are being taken seriously.

In reports: add one line under every important number stating its source, period, and limitations. The habit takes two minutes but prevents weeks of misreading.

In research: write your list of assumptions and limitations before you write your conclusions, not after. Conclusions written after facing the limitations honestly are almost always sturdier.

At home and on social media: before sharing a surprising chart or number, stop for ten seconds and look for the original source. Most viral numbers aren’t wrong; they just lose their context along the way.

A note on sources and how to read this recap

The crime figures in this recap come from BPS Crime Statistics 2024 and were checked against the publication when each original article was written. The 85% model accuracy, salary, and questionnaire examples are illustrations to explain concepts, not data from real cases or specific clients. If you quote any number from this recap, take it from the original source and read its definition, exactly as we discussed.

What we’ll open next week

Next week the theme shifts to young people and the world of work, ahead of Youth Pledge Day on 28 October. We’ll look at how youth unemployment compares with the national figure, how unemployment breaks down by education level, and how popular claims about a whole generation can be tested with data. Midweek, we’ll also tackle an important methods question: why social media polls and random surveys can produce very different numbers.

As always, we start from the official numbers, then from the questions. If there’s a figure about young people or jobs that confuses you, send it over. I’ll add it to the discussion.

Five reflection questions for the weekend

1. Which number did you quote most often this week, and do you know how it’s calculated?
2. Is there a decision at your workplace that was made from a single average without looking at the spread?
3. Who doesn’t show up in the data you use?
4. Which analysis method did you pick because “that’s how it’s usually done,” rather than because it fits the question?
5. If your number is wrong, who would be hurt most, and how could you find out earlier?

If you can only take one lesson

If I had to pick just one, my choice is the same as in previous weeks: before you believe a number, ask what was counted and what was left out. The seven lessons above are just variations of that question, applied to different data.

Tomorrow we start a new week with a topic close to many of us: young people and work, ahead of Youth Pledge Day. Same pattern: official numbers first, then questions.

Of the seven lessons above, which feels most relevant to your work or research? Share it in the comments. I always enjoy reading examples from the field.

Read also: need help analyzing and interpreting your research or business data? See our data analysis and statistical services.

---

This morning I answered questions about sample size, normality, and p-values. This afternoon it’s the turn of the questions that come up every time people read crime statistics. Because of my background in criminology, I often get contacted by people who aren’t studying criminology at all: journalists, students from other majors, local government staff, even parents choosing where to live.

The four questions below cover almost everything I’ve ever been asked. The answers draw on BPS (Statistics Indonesia) Crime Statistics 2024.

1. “Which city or regional police area is the safest?”

It sounds like a simple question, but the answer depends on what you mean by “safe.”

If you look at the number of incidents, Metro Jaya Regional Police (Greater Jakarta) ranks highest with 77,261 incidents in 2024, followed by North Sumatra (60,724) and East Java (60,102). But those totals don’t account for population. Using the crime rate (crimes per 100,000 people), the risk ranking becomes North Sulawesi at 540, West Papua at 501, and South Sulawesi at 440, while Metro Jaya drops to 8th place at 316.

So when comparing regions, use a per-population ratio, not raw counts. And remember that the crime rate itself is calculated from recorded incidents, so each region’s reporting rate shapes it too. A region with easy-to-use reporting services can look “more dangerous” precisely because its residents report more often.

2. “Drug cases rose 12%. Does that mean drug trafficking is increasing?”

Recorded narcotics cases rose from 39,496 to 44,295. That fact is correct. The conclusion that “trafficking is increasing” doesn’t necessarily follow.

Narcotics offenses are crimes without a reporting victim. Nobody walks into a police station to report themselves as a buyer. Cases get recorded when officers find, arrest, or investigate. That’s why the numbers depend heavily on enforcement intensity. Step up the operations and the numbers rise. Scale them back and the numbers fall, while the actual trade may stay exactly the same.

The more interesting research question is: what changes that enforcement intensity, and is there comparison data, such as drug-use surveys or rehabilitation service records, that could test our interpretation? For students, this is a rich thesis topic.

3. “Why can police data and survey data be so different?”

The clearest example is West Java. Police data puts it as the province with the lowest crime rate (89 per 100,000), yet the Susenas household survey shows that 1.30% of its residents say they have been victims of crime, the third highest in the country and well above the national average of 0.73%.

Those two numbers don’t cancel each other out. They measure different things in different units. Police data counts reported incidents. The survey counts the percentage of residents who say they were victims, regardless of whether they reported it. The gap is valuable information, not an error: it shows how far official records sit from what people actually experience.

A practical way to compare them is to use rankings rather than raw numbers, then compute the rank correlation between the two measures. If the rankings move in the same direction, both sources “agree” on which regions are more prone to crime. If they move in opposite directions, something needs explaining, such as differences in reporting rates.

4. “What is the dark figure of crime?”

The dark figure of crime is the gap between the crimes that actually happen and the crimes that get officially recorded. Criminologists have used the term for a long time, because it was clear from the start that police records only capture part of reality.

The data is right there in Crime Statistics 2024: only about 20.28% of crime victims reported to the police, down from 23.36% the year before. As a very rough illustration, if a type of crime has 1,000 recorded cases and only 20% of victims report, the actual number of victims could be around 5,000. That calculation rests on many assumptions, such as an identical reporting rate across all regions, which is clearly not true. But it’s enough to show the scale.

Summary of answers

Question	Short answer
Which region is the safest?	Compare per 100,000 people, not raw counts
Drug cases up means trafficking up?	Not necessarily: the number also measures how intense enforcement is
Why do police and survey data differ?	They measure different things: reported incidents vs. residents’ experience
What is the dark figure?	The gap between actual and recorded crime; about 80% of victims didn’t report, per the 2024 survey

Bonus question: “Are big cities always more dangerous?”

This question usually follows. The criminological answer has layers. Big cities do offer more opportunities for crime: more people, vehicles, and valuables in the same place, plus looser social control than small neighborhoods where everyone knows each other. On the other hand, big cities also tend to have more accessible reporting services, so more incidents get recorded.

As a result, recorded figures in big cities can be high for two reasons at once: more incidents, and easier reporting. Untangling how much each contributes is one of the biggest research challenges in this field, and it’s why researchers are rarely satisfied with a single data source.

Mini exercise: calculate a crime rate yourself

A crime rate is actually a simple calculation: the number of incidents divided by the population, multiplied by 100,000. Here’s an illustration with made-up numbers, just to show the logic.

Suppose Region A records 3,000 incidents with a population of 600,000, and Region B records 6,000 incidents with a population of 2,000,000. Region B has twice as many cases. But Region A’s crime rate is 3,000 ÷ 600,000 × 100,000 = 500, while Region B’s is 6,000 ÷ 2,000,000 × 100,000 = 300. Per resident, Region A is actually the riskier one.

Try it again with data from your own region. Don’t forget to note the year, the source of the population figures, and whether every type of incident is included.

If you want to explore the data yourself

The Crime Statistics publication is available on the BPS website. The most practical way to start: open the latest edition, read the introduction and technical notes first (that’s where definitions and data sources are explained), and only then look at the tables. Compare one table from police registration data with one table from the survey (Susenas), then write down three things: what is being measured, which unit is used, and who isn’t counted. The exercise takes about an hour and sharpens your reading enormously.

A word of caution

Crime statistics are about real people: victims, families, and offenders. When we talk about “numbers,” we’re talking about events that are sometimes painful. That’s why our language should be careful, not sensational, and never blame victims. Good statistics help us understand and prevent, not frighten.

If you have a question about crime statistics you’d like explored in more depth, leave it in the comments.

Read also: Ask a Statistician: 3 Questions I Get Asked the Most and 5 Misconceptions About Official Statistics.

Category: Data literacy · Series “Reading Indonesian Numbers” #1 · Source: BPS, Crime Statistics 2024

There is a sentence I often hear in boardrooms, social media debate threads, and academic thesis defense hearings: “But this is official data.”

That phrase is frequently swung like a hammer. The moment it is spoken, the discussion is treated as settled.

Yet official data is like a grocery scale: it is usually accurate, but it only weighs what you place onto it. If you only weigh a fraction of what is in the shop, the reading is still “correct”, but it tells you nothing about what fills the rest of the shelves.

Today, I am launching a new series: “Reading Indonesian Numbers.” Each edition will take one or two official statistics, unpack what they actually measure, and leave you with a question you can apply to any other dataset. No advanced statistical background is required—just genuine curiosity and a bit of patience to read the footnotes. The goal is simple: building the habit of asking one essential question before believing any figure: what is this number actually measuring?

To kick off, I have selected five of the most frequent misconceptions. The examples come from the 2024 Crime Statistics published by Statistics Indonesia (BPS), because that is where the analytical pitfalls are most glaring.

Misconception 1: “Official figures represent the full reality”

Let’s begin with the absolute foundation. The crime statistics you see on the news almost always originate from police records, meaning incidents that are formally reported and documented. Incidents that go unreported never enter the ledger.

How large is the gap? BPS has another way to look beneath the surface: the National Socioeconomic Survey (Susenas) asks citizens directly whether they have been victimized and whether they reported it to law enforcement. The result: in 2024, only about 20.28% of crime victims reported the incident to the police. A year earlier, that figure was 23.36%.

This means police registries only capture roughly one-fifth of what citizens actually experience. The figures remain official and valid as records of formal reports. The error lies in treating them as the total volume of actual occurrences. In criminology, this disparity has a well-known name: the dark figure of crime.

Misconception 2: “A rising number means things are getting worse”

An increase is a mathematical fact. But why it increased is another question entirely, and the answer is rarely as alarming as newspaper headlines suggest.

Take narcotics cases. Recorded offenses rose from 39,496 in 2023 to 44,295 in 2024, an increase of roughly 12%. At first glance, it sounds as if drug distribution surged dramatically. However, narcotics is an offense category that virtually lacks citizen reporters. Both buyers and sellers are willing participants; victims do not walk into police stations to lodge complaints. New cases are only registered when law enforcement officers uncover them through proactive operations.

Therefore, the tally can rise because trafficking increased, or simply because enforcement crackdowns intensified. A simple analogy: if you double the number of health inspectors in a local market, the number of recorded sanitation violations will inevitably climb, even if market conditions have not deteriorated at all. The figure measures how aggressively authorities are looking, not solely how much is happening.

Misconception 3: “Regions with the most cases are the most dangerous”

This is a classic trap that routinely produces misleading headlines. In 2024, the Greater Jakarta Metro Police (Polda Metro Jaya) recorded 77,261 incidents—the highest nationwide, accounting for roughly 13.75% of the national total. It was followed by North Sumatra with 60,724 cases and East Java with 60,102.

However, raw case counts fail to account for population size. A populous territory will almost always generate large absolute numbers. To compare regions fairly, BPS uses the crime rate, which measures recorded crimes per 100,000 residents. When calculated this way, the ranking shifts completely: North Sulawesi tops the list at 540, West Papua at 501, and South Sulawesi at 440. Metro Jaya, the raw volume leader, sits in 8th place with a crime rate of 316.

Asking “which region has the highest number of incidents?” and “which region poses the greatest risk to an individual resident?” are two completely different questions with two completely different answers.

Misconception 4: “Total crimes equal actual criminal offenses”

In 2024, BPS recorded a national total of 561,993 incidents. That is a massive number, easily weaponized to stoke public anxiety. But look closely at the components. The category of offenses against public order and state security includes road traffic accidents—and traffic accidents make up approximately 27% of the entire national total.

This does not mean the statistic is inaccurate. It is a matter of administrative classification: BPS follows the police record system, where traffic collisions are formally docketed. However, when readers encounter the phrase “561,993 crimes”, they picture burglaries, assaults, or fraud, not car crashes on the highway. Asking one straightforward question—”what exactly is lumped into this total?”—instantly clarifies how the figure should be interpreted.

Misconception 5: “A single average represents everyone”

Finally, the ubiquitous average. In the BPS national survey, approximately 85.47% of households reported feeling safe walking alone in their neighborhood at night. That figure sounds comforting, and as a national average, it is.

Yet look at the geographic dispersion. Central Java recorded 91.92%, DI Yogyakarta 91.37%, and East Java 90.38%. At the opposite end, Central Papua stood at just 53.50%, Gorontalo at 69.02%, and Southwest Papua at 69.15%. The gap between the highest and lowest provinces spans nearly 40 percentage points.

A national average is like the average temperature of a continent: useful for bird’s-eye overviews, but useless if you are deciding whether to pack a coat for the mountains. If you are a policymaker or a business leader managing branches nationwide, relying on blanket averages is the quickest way to overlook acute regional vulnerabilities.

Bonus: When definitions change over time

There is another pitfall that rarely makes top-five lists but frequently distorts perception: comparing longitudinal trends across years without verifying whether definitions remained stable. The national incident total reported by BPS hovered around 239,000 in 2021, before jumping to approximately 585,000 in 2023 and 562,000 in 2024. A leap of that magnitude is far too abrupt to be explained solely by shifts in criminal behavior. Much of it stems from expanded recording coverage and structural reclassification of incident types.

Whenever you encounter a sharply ascending trendline, check the methodological notes. Ask: did definition and scope remain consistent across the timeline? If not, the trendline must be interpreted with extreme caution, or separated into distinct periods that are not directly compared.

Summary in a single table

Misconception	What it actually measures	Example from 2024 BPS Data
Official figures = complete reality	Measures reported incidents	Only ~20.28% of crime victims report to police
Rising count = worsening conditions	Can reflect intensified enforcement	Narcotics cases rose from 39,496 to 44,295
Highest count = most hazardous area	Must be normalized per population	Metro Jaya has 77,261 cases, but ranks 8th in crime rate
Total count = purely criminal offenses	Broad statistical definitions	~27% of 561,993 recorded incidents are traffic accidents
One average fits all	Wide regional variation	85.47% feel safe nationally, but only 53.50% in Central Papua

One habit that changes how you read data

All five misconceptions trace back to a single root cause: we consume numbers without asking who counted them, what was included, and what was left uncounted. Those three questions alone will shield you from the majority of hasty and erroneous conclusions.

This is not about distrusting official agencies. Reputable institutions openly document definitions and analytical boundaries in their publications. The reader’s responsibility is to actually read them, rather than simply quoting the headline number.

Next week, we will explore datasets closer to daily life: youth demographics, employment trends, and generational claims. The framework remains identical: examine the official figure, then ask the right questions.

What about your experience? Which of these five statistical misconceptions do you encounter most often at work, university, or in family discussion groups? Share your thoughts below.

Also explore our Data Analysis & Statistical Services for research design, survey methodology, and quantitative data analysis consulting.

---

We usually picture corruption as something huge: billions in value, senior officials, a courtroom full of cameras. But there is another side to corruption that is far more everyday, and precisely because it is everyday, it slips under the radar: extra money to speed up paperwork, a “thank you” that is practically demanded, or a little grease so things don’t get complicated.

The 2024/2025 Crime Statistics from BPS (Statistics Indonesia) include two measures that spotlight this side, drawn from the Anti-Corruption Behavior Survey (SPAK). Let’s look at what the data says, and what it doesn’t.

Two indicators from SPAK

1. The Anti-Corruption Behavior Index (IPAK). It runs from 0 to 5. The closer to 5, the stronger the anti-corruption behavior in society. The closer to 0, the more tolerant (permissive) people are toward corruption. The result: IPAK fell from 3.92 in 2023 to 3.85 in 2024, a drop of 0.07 points. IPAK is built from two dimensions: perception (how people judge habits in their family, community, and the public) and experience (direct experience of accessing public services).

2. The share of people who dealt with public services and handed over money, goods, or favors outside the rules. It rose from 16.33% in 2023 to 17.17% in 2024. BPS flags this as an important note because it points to growing permissiveness toward small-scale corruption, and to gaps in the public service system that can be exploited, whether on the citizen’s initiative or at an official’s request.

Put simply: roughly one in six people who deal with public services say they have given something outside the rules.

What is measured, and what is not

As always, let’s check how it was measured.

• What is measured is petty corruption: bribery, extortion, and nepotism in everyday public services. BPS states explicitly that the survey does not cover grand corruption. So IPAK says nothing about corruption at the policy level, in big procurement, or in abuse of power by senior officials.
• The sample is limited. In 2024, SPAK was carried out in 186 regencies/cities across 38 provinces with about 11,000 households as respondents. Results can only be presented at the national level, not broken down by province or region.
• It rests on admissions and perceptions. Admitting to paying outside the rules is a sensitive topic. Respondents may hold back out of shame or fear, so the figure may be lower than reality (social desirability bias). Or the reverse: some respondents may interpret “extra fees” differently.
• Giving is not the same as being asked. The indicator mixes two situations: citizens who offer on their own, and citizens who are asked or pressured. From a single 17.17% we cannot separate them. Yet for policy they differ: the first is about a permissive culture, the second about officials’ behavior and oversight systems.

Two datasets moving differently

It is interesting to compare SPAK with police incident data. In the Polri registration data, recorded corruption cases rose slightly from 483 to 512 in 2024. BPS says this increase deserves special attention because corruption, as white-collar crime, causes large-scale losses.

The two figures cannot be compared directly. 512 is the number of corruption cases handled and recorded by police (generally cases that have drawn attention and met the requirements for enforcement). 17.17% is the share of citizens who say they experienced petty bribery. The scale, definitions, and sources are very far apart. But that is exactly why the comparison teaches something: what citizens experience daily and what authorities handle are two entirely different layers of the same problem. Many small payments will never become criminal cases, which is why surveys like SPAK are needed to see the layer that goes unrecorded.

Why small-scale corruption still matters

It is tempting to wave off small payments as “not much”. From a criminology and public policy standpoint, petty corruption has several serious effects:

• Normalization. When “grease money” is seen as normal, the line between what is acceptable and what is a violation blurs. IPAK itself measures this permissiveness, and its decline is a signal that tolerance has crept up.
• Unfair burden. Small payments weigh more on low-income groups, who depend most on public services and have the fewest alternatives.
• Erosion of trust. Paying “extra” makes people trust institutions less, which in turn reduces willingness to report, as we see in the low reporting rate for crime.
• A doorway to bigger corruption. A culture that accepts small grease payments is fertile ground for larger practices.

What can be done?

The data itself offers no solution, but it points to several policy questions:

1. Which services are most vulnerable? A breakdown by service type (police, document processing, education, health, licensing) needs to be analyzed so interventions hit the right target.
2. Does digitizing services reduce face-to-face contact that opens room for payments? This is a hypothesis that can be tested with data.
3. How effective are complaint mechanisms? The low reporting of crime in general shows that complaint channels must be safe and trusted.
4. How does education affect permissiveness? IPAK can be tracked year to year to assess impact.

For researchers and students

SPAK and IPAK data are interesting research material, for example: what factors influence permissiveness toward petty corruption, how perception relates to actual experience, and how IPAK trends move from 2012 to 2024. Be careful with the limitations already discussed: national-level only, based on admissions, and covering only small-scale corruption.

How do you measure sensitive topics more honestly?

Surveys about bribery face a classic problem: respondents are reluctant to admit behavior seen as improper. Researchers have several techniques to reduce this bias. First, indirect questions, such as asking whether “people around you” have done it. Second, the randomized response technique, where respondents answer under a random rule that protects their privacy (for instance flipping a coin to decide whether to answer truthfully or give a set answer), so the researcher does not know any individual’s answer but can still estimate the overall proportion. Third, the list experiment, where one group sees a list of four statements and another sees five (the same four plus one sensitive item), and both only report how many statements apply to them. The difference in averages between the two groups estimates the share admitting the sensitive behavior, without any respondent having to state it directly. These techniques are more complex but worth considering for academic research on sensitive behavior.

Quick Q&A

Does 17.17% mean one in six public officials is corrupt? No. The figure shows the share of citizens who admit giving something outside the rules, and not all of them were necessarily asked by an official. It also says nothing about how many officials were involved.

Is the drop in IPAK from 3.92 to 3.85 big? A 0.07 gap on a 0 to 5 scale looks small, but the downward direction relates to rising permissiveness. To judge whether it is meaningful, we need to look at the IPAK series over several years and account for the survey’s sampling error.

Read also: [Market research and survey services](https://datalyticsindo.com/en/market-research-survey/)

---

Source: BPS (Statistics Indonesia), Criminal Statistics 2024/2025 (police records and the Susenas household survey)

Note: this article discusses sexual violence and violence against children. It presents data, handled with care.

There is one sentence I keep repeating in my statistics classes: “A total that falls can hide something that rises.” Indonesia’s Criminal Statistics 2024/2025 gives a sad but important example of it.

The total number of recorded crimes fell from 584,991 in 2023 to 561,993 in 2024. If we stopped there, we might feel relieved. But inside that total, some categories moved the opposite way, and they concern the most vulnerable groups.

Crimes against decency: the total went up

BPS notes that crimes against decency, as a whole, rose in 2024. The category covers four types: rape, sexual abuse (molestation), sexual violence, and sexual intercourse with a child. The breakdown:

Type	2023	2024	Direction
Sexual intercourse with a child	3,691	4,525	up about 23%
Sexual violence	1,410	1,861	up about 32%
Sexual abuse (molestation)	2,739	2,004	down
Rape	1,230	1,029	down

According to BPS, the rise in the total was driven by sharp increases in sexual intercourse with a child and in sexual violence. The regional police commands with the most cases in 2024 were North Sumatra (750), East Java (708), and South Sulawesi (608).

Domestic violence rose too

Among crimes causing physical injury, assault declined (from 51,106 to 48,622). Two other types went up, though: group assault (from 16,441 to 17,078) and domestic violence (from 10,783 to 11,524, up about 7%).

The other side of the data: who reports, and who does not?

The Susenas survey adds another view of vulnerability.

Women. Women made up about 37.84% of all crime victims in Susenas 2024, up from 36.83% a year earlier. BPS stresses that women tend to face different kinds of crime, especially sexual and domestic violence, which is why gender-based analysis matters. When it comes to reporting violence, the share of female victims who went to the police rose from 38.10% to 41.21%, while for men it fell from 56.05% to 37.84%. In other words, in 2024 female victims reported more often than male victims, the reverse of 2023.

Children. Of all crime victims who reported, children were only 4.24% in 2024 (down from 7.08%). BPS points to possible reasons: limited knowledge of their legal rights, dependence on adults, and fear of perpetrators who may be people close to them. It cites UNICEF’s finding that most violence against children goes unreported because it happens within the family or at school.

People with disabilities. The share of crime victims with disabilities rose from 1.70% to 1.78%, and BPS cites UNODC’s observation that people with disabilities are more vulnerable to violence and exploitation.

How should we read these increases?

This is where care is needed. There are two broad explanations. They are not mutually exclusive, and registered crime figures alone cannot tell them apart:

First possibility: incidents really did increase. That is bad news, and it calls for stronger prevention.

Second possibility: reporting improved. Sexual crimes and domestic violence have a very large “dark figure”, because victims are often afraid, ashamed, or dependent on the perpetrator. If complaint services become easier to reach, victims grow braver, and officers respond better, the number of recorded cases will rise even if actual incidents do not. Indonesia has Law No. 12 of 2022 on Sexual Violence Crimes, meant to strengthen victim protection and the prosecution of perpetrators, and BPS mentions it in this context. Whether the law affected reporting in 2024 is a reasonable hypothesis, but it needs to be tested, not assumed.

The most responsible stance: take both seriously. If reporting improved, that is a good sign, but it means a long-hidden problem is only now becoming visible. If incidents rose, the urgency is even higher. Either way, the message is the same: protection and recovery for victims must be strengthened.

What this means for how we read and talk about this data

First, do not use the numbers for sensationalism. This topic involves real victims, often children. Avoid exaggerated headlines, and present the context (such as reporting effects) alongside the figures.

Second, mind the unit of analysis. Police data count incidents, not victims. One incident can involve several victims, and one victim can experience more than one incident.

Third, do not infer causes from aggregate data. A rising number does not explain why. Causes need deeper qualitative and quantitative research: offender patterns, locations, family and environmental risk factors, and how well protection services work.

Fourth, remember the victims behind the numbers. Every figure in the table is a person. Good statistics are not an excuse to distance ourselves from humanity; they are a tool to make policy reach those who need it most.

If you or someone around you needs help

If you know of or are experiencing violence against women or children in Indonesia, official complaint services such as SAPA 129 from the Ministry of Women’s Empowerment and Child Protection, as well as local service providers in your area, can be the first door to support. In an emergency, contact local authorities.

A closing note for researchers

This topic needs more careful and ethical research: multi-year trend analysis, separating reporting effects from incidence effects, evaluating protection services, and studies that involve victims with strict privacy safeguards. If you are considering this topic for a thesis or dissertation, make research ethics and respondent protection part of the design from the start.

From data to prevention: what can be done?

Descriptive data like this does not explain causes, but it helps choose a direction. Evidence-based prevention of violence against children and women generally works on several levels. In the family: positive parenting and support for vulnerable households. In schools: age-appropriate education about the body, boundaries, and how to report, plus teachers who know how to respond. In the community: safe, easy complaint channels and attentive community leaders. In the system: integrated services for victims (medical, psychological, legal) so that reporting does not mean an exhausting process. Every program needs clear evaluation indicators, such as changes in reporting rates, service response times, and survivor satisfaction, not just the number of activities held.

Ethics in reporting and writing about data

For anyone presenting this data, here are the principles I hold to. Do not name victims or give clues that could identify them, especially children. Avoid language that blames victims or normalizes violence. Do not use numbers as a tool for sensation. Include information about support services. And explain that rising numbers may also mean more people dare to report, without using that as a reason to downplay the problem. Data about children and women demands more care than data on property crime, because behind it are deeply vulnerable people.