Kategori: Statistics Education

In recent years, I have increasingly heard remarks like this in meetings: “We ran an Instagram poll yesterday, and 80% of our followers agreed.” Or: “Our customer WhatsApp group says they love the new product.” Such polls are quick, inexpensive, and feel inherently democratic. Anyone is welcome to participate.

That, however, is precisely where the trouble begins. Because anyone can participate, you never truly know who actually responded. And in survey research, who answers is almost always far more critical than how many answered.

This article examines the fundamental distinction between social media polls and random probability surveys, highlights a famous historical lesson, and provides a practical framework for choosing the right methodology for your decisions.

The Core Difference: Who Selects the Respondents

In a social media poll, respondents select themselves. Only individuals who happen to see the post, feel engaged by the topic, and are willing to spend the time will respond. In research methodology, this is called a self-selected sample.

A few defining characteristics of self-selected samples deserve close attention:

• Coverage is restricted to followers. An account’s followers do not represent the general population. They tend to share similar interests, demographics, or viewpoints with the account owner.
• Stronger opinions are overrepresented. People with intense enthusiasm or vehement disapproval are far more motivated to participate than those with moderate views.
• Algorithms shape visibility. Who sees the post is governed by platform algorithms, not the researcher’s design.
• A single individual can vote repeatedly across multiple accounts or devices.

In a random probability survey, the researcher selects respondents using probabilistic procedures. Every member of the defined population has a known, non-zero chance of selection. Because that probability is known, researchers can mathematically calculate how far the sample estimate might deviate from the true population value—what we call the margin of error.

Why Big Numbers Cannot Rescue a Biased Sample

Many assume that a poll with 10,000 votes must be more reliable than a survey of 400 respondents. The intuition seems straightforward: larger numbers must mean greater representation.

That reasoning holds true only for random probability samples. Increasing sample size reduces random sampling error. But if the selection mechanism itself is biased, collecting more responses does not reduce systematic error (bias). It merely produces an inaccurate estimate with greater statistical confidence.

A simple analogy illustrates this: if your bathroom scale is miscalibrated by 2 kilograms, stepping on it a hundred times will not make the weight reading any more accurate. The average remains 2 kilograms off; you are simply more confident in the wrong number.

A Historical Lesson: The 1936 Literary Digest Poll

The most renowned cautionary tale in survey history unfolded during the 1936 United States presidential election. The Literary Digest, a wildly popular magazine at the time, mailed millions of mock ballots to its subscribers and to individuals listed in telephone directories and automobile registries. Over two million ballots were returned.

Based on this massive response, the Literary Digest confidently projected that Alf Landon would defeat incumbent Franklin D. Roosevelt by a decisive margin. The actual election yielded the exact opposite: Roosevelt secured a historic landslide victory.

What went wrong? Analysts later highlighted two fatal flaws. First, the contact lists drew disproportionately from affluent households during the height of the Great Depression. Second, those who returned the postcards were not a random cross-section of recipients, introducing severe nonresponse bias. In the very same election cycle, pioneering pollsters using vastly smaller samples—selected through disciplined quota and probability sampling—correctly predicted Roosevelt’s triumph.

The lesson has endured for nearly a century: bigger is not synonymous with better.

Margin of Error: What Can and Cannot Be Calculated

For a simple random sample, the margin of error for a proportion at a 95% confidence level can be calculated using standard formulas. Under the most conservative assumption (a 50% proportion) in a large population, the numbers look like this:

Two observations stand out. First, cutting the margin in half (from roughly ±5% to ±2.5%) requires roughly four times the sample size. Precision comes at a real cost. Second, self-selected polls have no mathematically valid margin of error because there is no underlying probability mechanism.

A vital caveat: margin of error only captures sampling error. Even random surveys encounter non-sampling errors, such as incomplete sampling frames, non-response attrition, and leading question wording. A sound survey is never completely free of error, but it understands where vulnerabilities lie and systematically works to minimize them.

Are Social Media Polls Useless?

Not at all. Social media polls remain useful for specific, targeted purposes:

• Taking the pulse of your own audience, such as learning which content themes your followers prefer.
• Sparking community engagement and discussion.
• Generating exploratory hypotheses prior to investing in formal research.

The misstep occurs when organizations use casual poll results to declare what “the broader public” or “the market” demands, and then make significant capital or strategic bets based on those claims.

A Decision Framework for Choosing Your Method

Before deciding between an informal poll and a formal survey, ask yourself three core questions:

1. Whose opinion am I truly trying to measure? If the answer is “the people following my account,” a quick poll suffices. If the answer is “consumers across Metro Jakarta” or “all employees nationwide,” you require a representative sample.
2. What is the magnitude of the decision? For small, easily reversible choices, rough directional feedback is often adequate. For product launches, capital investments, or organizational policy, you need defensible data.
3. Do I need to know how certain the findings are? If yes, only probability-based sampling can provide measurable statistical bounds.

What About Online Surveys?

A common follow-up question is whether all online surveys suffer from the same weaknesses as social media polls. The answer is no. Online surveys can be methodologically rigorous—for instance, by sending survey invitations exclusively to randomly sampled customer or employee lists, utilizing verified probability-recruited research panels, or applying post-stratification weighting to align sample demographics with population parameters.

What matters is not the medium—digital versus face-to-face—but the underlying respondent selection procedure. An online survey sent to a random sample of an employee roster is vastly more trustworthy than an intercept survey conducted on a street corner where interviewers pick respondents at convenience.

Closing Thoughts

In today’s digital landscape, collecting thousands of responses takes just a few clicks. Yet because collecting data has become so effortless, critical thinking is more essential than ever. Thousands of self-selected responses can easily be far less accurate than a few hundred properly sampled ones.

Before trusting any survey finding, the most valuable question to ask remains unchanged: how were the respondents selected?

Read also: planning a customer survey, brand research, or market study with rigorous, representative sampling? Explore our Market Research & Survey Services.

Every time I run a training session or supervise a thesis, the same three questions come up. The wording varies, but the core is identical. This Saturday, I’m answering all three in one place, in language you can use right away.

The questions: how large a sample is enough, does non-normal data require a non-parametric test, and does a non-significant result mean your research failed. Let’s start with the one I hear most often.

Question 1: “How large a sample is enough?”

My honest answer: it depends on what you want the sample to tell you. There is no magic number.

If your goal is estimating a proportion (for example, “what percentage of customers are satisfied?”), sample size determines precision, measured by the margin of error. At a 95% confidence level and the most conservative assumption (proportion of 50%), the required random sample is roughly:

Notice the pattern: to cut the margin from ±5% to ±2.5%, you don’t just double the sample—you need roughly four times as many respondents. That’s why tight precision is expensive.

Two important caveats. First, these numbers assume a random sample. A convenience sample of 10,000 from a group chat poll isn’t automatically more accurate than 400 randomly selected respondents, because what randomization guarantees is the ability to quantify the error, not the sheer count. Second, for small populations the requirement drops. With a finite-population correction, a population of 2,000 with a ±5% margin needs only about 323 respondents, and a population of 500 about 218.

If your goal isn’t estimating a proportion but testing relationships or building a model (regression, SEM, PLS-SEM), the logic changes: what matters is statistical power, the number of variables or paths in the model, and the effect size you want to detect. So don’t copy “100 respondents” from someone else’s thesis without checking whether their objective matches yours.

Question 2: “My data isn’t normal. I have to use a non-parametric test, right?”

This is the question that sends students into a panic two weeks before their defense. My short answer: not necessarily, and you may be testing normality on the wrong thing.

In many analyses—regression in particular—the normality assumption applies to the residuals (the difference between observed and predicted values), not to each raw variable individually. Raw variables can be skewed, yet the residuals may still satisfy the assumption.

A few things to check before switching to a non-parametric test:

• What does your chosen test actually assume? The t-test and ANOVA, for instance, require normality within each group or of the residuals, and both are reasonably robust against mild violations when the sample isn’t tiny.
• Look at the visuals. Histograms and Q-Q plots are often more informative than a single p-value from a normality test. Formal tests like Shapiro-Wilk can “reject” normality in large samples due to deviations that are practically trivial.
• Know your alternatives. If the assumption truly fails, non-parametric tests aren’t the only option. You can also transform the data or use methods that are robust to violations.

The wording in your results section matters too. Instead of writing “the data were not normal so a non-parametric test was used,” examiners appreciate a sentence that states which assumption was tested, with which tool, on which object, and what the consequence was.

Question 3: “My result isn’t significant. Does that mean my research failed?”

No. And this is probably the most damaging misconception in research.

A p-value answers a narrow question: if there really were no effect, how rarely would I see data as extreme as this? It does not answer “is the effect important,” and a large p is not evidence that the effect doesn’t exist. It could simply mean your sample was too small to detect it.

Imagine a training program that raises the average competency score by 2 points on a scale with a standard deviation of 10 (a small effect). Tested on 30 people, that effect will almost certainly go undetected, and you might conclude “the training had no impact.” Tested on 3,000 people, the exact same effect produces a tiny p-value and you conclude “significant impact.” The effect is still two points. Only the sample size changed.

That’s why I recommend the following when reporting results:

1. Report the effect size (e.g., the coefficient or the mean difference), not just p.
2. Include a confidence interval so readers can see the plausible range.
3. Discuss the sample’s power to detect the effect, especially when the result is non-significant.
4. Ask about practical relevance: does a difference that size matter for decisions or costs?

A non-significant result, reported honestly, is a legitimate finding. What’s dangerous is the habit of trying different analyses until something is significant, then reporting only that one.

Bonus: three sentences you can adapt for your methods and results sections

Sometimes what trips students up isn’t the analysis but the writing. Here are template sentences you can tailor to your own case. Fill in the numbers with your own data and decisions.

On sample size: “The sample size was determined to achieve a maximum margin of error of ±5% at a 95% confidence level, requiring at least approximately 385 respondents for a large population. The sample was selected using [state the sampling technique].”

On assumptions: “The normality assumption was assessed on the model residuals using a Q-Q plot and the [state the test], and the results [met or did not meet the assumption]. Accordingly, the analysis proceeded with [state the method].”

On a non-significant result: “The effect of [variable A] on [variable B] was not statistically significant (coefficient [value]; 95% confidence interval [lower] to [upper]). With a sample size of [n], this study had limited power to detect a small effect, so this result should be interpreted as [an honest interpretation] rather than as evidence of no effect.”

Note that all three sentences state the reason, not just the decision. That’s what examiners look for: proof that you know why you did something.

The common thread

All three answers share one principle: don’t apply rules of thumb without understanding the rationale behind them. Sample-size thresholds, normality tests, and the p < 0.05 cutoff are all tools. Tools used without understanding their purpose produce conclusions that look neat but are hard to defend in a thesis defense or a boardroom.

If you have another question, leave it in the comments or send a direct message. Good questions will be featured in the next Saturday edition.

Read also: [Data Analysis & Statistical Services](https://datalyticsindo.com/en/data-analysis-statistical-services/) – statistical analysis, data processing, and methodology support from the Datalytics team.

When you picture crime, what usually comes to mind is theft, fraud, or violence. Few people immediately think of forests cleared without permits, fish poached from the sea, or mines operating outside the law. Yet crimes against the environment cause enormous damage, and their victims include not just people living today, but future generations as well.

The 2024/2025 Crime Statistics from BPS (Statistics Indonesia) include a dedicated section on this category of crime. Inside it, one figure stands out prominently: North Sumatra Regional Police recorded 1,334 incidents, roughly 32% of all environmental crimes in Indonesia. Let’s examine that number carefully.

What counts as an environmental crime?

BPS groups five types into this category:

1. Forest destruction prevention and suppression offences (illegal logging)
2. Plantation offences
3. Fisheries offences (illegal fishing)
4. Mineral and coal mining offences (illegal mining)
5. Oil and gas offences

The figures for 2024 compared to 2023:

The two types that increased—plantation and oil and gas offences—account for approximately 65% of all environmental crimes in 2024. Both also saw substantial growth compared to the prior year: plantation offences climbed about 26%, while oil and gas rose around 21%.

An uneven geographic distribution

The three regional police commands (Polda) with the highest incident counts:

• North Sumatra: 1,334 (about 32% of the national total)
• South Sumatra: 230
• East Java: 212

And those with the lowest:

• West Sulawesi: 1
• Papua: 11
• North Maluku: 19

The gap between North Sumatra and the second-ranked province is more than fivefold. Meanwhile, Papua, known for its vast rainforests and natural resources, recorded only 11 incidents.

The critical question: what does this number actually measure?

This is where critical thinking is essential. The question is not “where is environmental destruction greatest?”, but rather “where are environmental crimes most actively recorded and prosecuted?” Those are two completely different things.

First, these are crimes without reporting victims. Like narcotics, environmental crimes are rarely reported by individual victims. The victims are ecosystems, the broader public, or future generations. For a case to be recorded, law enforcement or inspectors must actively discover it, investigate, and take action. The recorded case count reflects enforcement and surveillance capacity, not simply the true underlying number of violations.

Second, an extremely low figure does not automatically mean a pristine area. Recording 1 incident in West Sulawesi or 11 in Papua could mean violations are genuinely rare. But it could also reflect limited surveillance across vast, rugged terrain with few enforcement personnel available. In such contexts, “unrecorded” is not the same as “non-existent.”

Third, a high figure can indicate active enforcement. A province with elevated numbers may indeed have more violations. But it may also have a police force that is particularly active in investigating and cracking down, bringing more cases to light. In certain settings, a rising number signals improving enforcement rather than deteriorating environmental conditions.

Fourth, the local economic structure shapes the types of offences. Plantation hubs will naturally record more plantation offences. Oil-producing regions will log more oil and gas offences. Regional disparities largely mirror differences in economic geography, not the moral character of residents.

The takeaway: enforcement statistics measure institutional activity

This is perhaps the most important lesson from this dataset, and it applies equally to narcotics, corruption, and many other crimes: police registration data for crimes without individual reporting victims measure law enforcement activity, not the true scale of the underlying problem. Using them to declare “Province X is the most degraded” or “Province Y is the cleanest” is a leap unsupported by data.

To measure actual environmental damage, we need different data sources: satellite imagery on forest cover change, licensing databases, field surveys, ecological loss assessments, and environmental quality metrics. Criminal statistics fill only one piece of the puzzle.

What can researchers explore further?

For researchers and students, several questions are worth exploring:

• Is there a relationship between recorded case counts and enforcement resources across regions (investigators, budgets, territorial reach)?
• How do environmental crime statistics correlate with remote sensing data on land cover change?
• Why did plantation and oil and gas offences rise while others declined? Were there regulatory shifts, commodity price changes, or specific operations?
• How do recorded environmental cases progress through prosecution, sentencing, and environmental remediation? Initial police statistics tell only the opening chapter of the legal process.

Each question demands additional data and a far more rigorous approach than simply contrasting headline totals across provinces.

Concluding thoughts

Environmental crime demands that we look beyond headline statistics. Data remains essential, provided we are candid about what it measures. The figure of 4,149 is not a balance sheet of environmental degradation in Indonesia. It is a record of cases that made it into the formal enforcement system. What remains unrecorded—and what has been degraded without documentation—may well be vastly larger.

Combining enforcement data with environmental indicators

To evaluate environmental crime comprehensively, enforcement statistics must be paired with independent environmental metrics. Remote sensing (satellite imagery) can track changes in canopy cover or illegal mining footprints. Concession and licensing registers document lawful activities. Water, air, and soil quality monitoring reveal real-world impact. Natural resource revenues show economic stakes. Laying these layers together allows us to ask a far more meaningful question: which areas show significant environmental disturbance yet report virtually no legal cases? That gap is where true surveillance deficiencies lie.

Community-based prevention

Law enforcement alone is rarely sufficient. Evidence from many regions indicates that environmental crimes are prevented most effectively when surrounding communities have a tangible stake in safeguarding resources—such as through community forestry, small-scale fisheries co-management, or secure whistleblowing channels that guarantee follow-up. Crime statistics capture these outcomes in an ironic way: recorded complaints may initially surge as community trust improves, signaling that the oversight system has finally begun to work.

Research ideas for students

Compelling research topics include: examining the link between provincial police enforcement capacity and recorded environmental case volumes; comparative analysis of environmental offence categories mapped to regional economic structure; or qualitative case studies examining how provinces with high reporting rates developed their specialized investigation units. Each pairs empirical rigor with institutional insight.

Quick Q&A

Why is the total number of environmental crimes relatively small compared to other offences? Partly because these offences are less frequently detected through citizen reports, and enforcement requires specialized resources (technical investigators, expert witnesses, formal damage assessments). Investigations are also technically complex and lengthy, meaning cases often take more than one reporting cycle to conclude.

Does a rising number mean environmental destruction is accelerating? Not necessarily. It can indicate strengthened enforcement and proactive inspections. This is why environmental crime records must always be interpreted alongside independent environmental condition indicators.

Read also: Data analysis and statistical services

One of the most interesting parts of Indonesia’s Criminal Statistics 2024/2025 does not come from police records or victimization surveys. It comes from the Village Potential Census (Podes), a BPS data collection that treats every village and urban ward (desa/kelurahan) as a unit of observation. It includes a question about mass brawls in the past year.

The result grabs attention: the share of villages and wards that experienced a mass brawl rose from 1.85% in 2021 to 2.41% in 2024. And the province with the highest share is… DKI Jakarta, at 18.73%.

Before jumping to the conclusion that Jakarta is the “most brawl-prone province”, let us read the numbers calmly. This is where statistical thinking helps us avoid rushing to conclusions.

What does Podes actually count?

Podes does not count the number of brawls. It counts the percentage of villages and wards that experienced at least one incident in the past year. So a ward with ten incidents and a ward with one incident are both counted as “experienced”. The indicator measures the spread of conflict, not its intensity.

The recorded types of mass brawl include several forms: fights between resident groups, between residents of different villages, clashes between residents and security forces, between residents and government officials, fights between students, and inter-ethnic conflict. In 2024 the most common type was fights between resident groups, which rose slightly from 1.00% (2021) to 1.09% (2024). BPS lists common causes such as insults, misunderstandings, juvenile delinquency, and old grudges.

National trend: up, but from a small base

The rise from 1.85% to 2.41% is a relative increase of about 30%. In absolute terms it is only about 0.56 percentage points. Both views are correct, and both should be stated together. Saying “up 30%” without context sounds frightening. Saying “only half a point” may downplay it. It is better to give both.

BPS interprets this as social conflict potential in the form of mass brawls still being fairly high and tending to spread to more areas, and names several possible factors: rising mobility, tension between residents, economic pressure, and local conflict resolution mechanisms that are not yet optimal. That is a reasonable explanation, but note that this data does not test causes. It only describes a pattern.

The reassuring side: most conflicts are settled peacefully. In 2024, 94.29% of incidents were resolved peacefully (down from 95.49% in 2021), and about 5.71% remained unresolved. Even a small decline deserves attention because it signals challenges in village-level resolution mechanisms.

Variation across provinces

The three provinces with the highest percentages in 2024:

And the lowest: Riau Islands (0%), South Kalimantan (0.20%), and North Kalimantan (0.41%). The gap between DKI Jakarta and the national average (2.41%) is almost eightfold. BPS says the difference reflects inequality in the intensity and frequency of mass conflict across regions, and points to socio-economic, cultural, and community interaction patterns as possible causes.

The statistical trap: areal units that are not comparable

This is where care is needed. The indicator is computed per village or ward. But villages and wards are not uniform in size. A ward in DKI Jakarta is typically densely populated with a small area. A village in Kalimantan or Papua may have few residents spread over a large area. In a small, dense ward, the chance of at least one incident in a year is naturally higher, because many people interact in a tight space. With smaller and denser units, the percentage of “wards that experienced an incident” tends to be higher even if the number of incidents per resident is not.

In statistics this relates to the modifiable areal unit problem: results can change depending on how we divide regions into units of analysis. The percentage of villages is not the same as incidents per 100,000 residents. They answer different questions.

So 18.73% for Jakarta does not automatically mean Jakarta residents are involved in mass brawls more often than residents of other provinces. What we can safely say: in Jakarta, the proportion of areal units (wards) that experienced at least one mass brawl in a year is the highest. To conclude more, we need measures adjusted for population and more detailed incident data.

What can we learn from a criminology perspective?

Mass brawls are interesting to study because they do not fit neatly into the “offender and victim” framework of, say, theft. They are collective crime, often rooted in community tension, identity, and rivalry. Some relevant theories: social disorganization theory (weak cohesion and informal control in dense neighborhoods with high turnover), strain theory (economic pressure driving frustration), and communal conflict approaches. The strength of peaceful resolution at village level (94.29%) also reminds us that local social capital, such as community leaders, resident mediation, and customary law, is an asset often neglected in security policy.

Three notes for readers of this kind of data

1. Separate spread from intensity. The percentage of areas with an incident is not the number of incidents.
2. Beware of areal units that are not comparable. Compare areas of similar size, or use per-capita measures.
3. Do not infer causes from patterns. Descriptive data points to where, not why.

Two interesting research designs

Podes data opens several research possibilities. The first is a multilevel model: villages and wards are nested within districts, regencies, and provinces. We can test village-level factors (density, inequality, facilities, neighborhood security activities) alongside broader regional factors, without violating independence assumptions. The second is spatial analysis: do mass brawl incidents cluster geographically? Techniques such as Moran’s I or spatial cluster analysis can show whether there are “hot spots” that cross administrative boundaries, so prevention policy can be designed across regions.

The role of early detection and mediation

BPS highlights the importance of strengthening village officials, community leaders, and community institutions in early detection and mediation. The fact that about 94% of conflicts end peacefully at the local level shows that this capacity already exists. A smart policy would strengthen the mechanisms already working, for example mediation training for village officials and communication forums between youth groups, rather than replacing them with a purely repressive approach. For fights between students, schools and families play an important role that police statistics do not capture.

Quick Q&A

Is a mass brawl the same as student brawling? Student brawls are one form. In the Podes classification, mass brawls also include fights between resident groups, between villages, clashes with security forces, and inter-ethnic conflict. Because the scope is wide, interpretation by type is more meaningful than the combined figure.

Is the rise from 1.85% to 2.41% alarming? It is a rise worth monitoring, but the base figure is still small and most cases are settled peacefully. A more useful question is whether the trend continues in the next census, and which regions show the sharpest increases.

Read also: [Data analysis and statistical services](https://datalyticsindo.com/en/data-analysis-statistical-services/)

Every time there is a general election or a local election, two things almost always happen. First, a polling firm announces its quick count a few hours after voting closes. Second, someone unhappy with the result says, “Quick count is a lie!” or, if the result suits them, “Quick count proved to be accurate!”

I have worked on political surveys and quick counts for several local elections, and one thing I always say: a quick count is not a prediction, not magic, and not just a number you can either trust blindly or dismiss as fake. It is a statistical method, and like any statistical method it has conditions for working well and limits we need to understand.

How does a quick count work?

The idea is simple. Instead of waiting for the official tally from every polling station (TPS), which number in the hundreds of thousands nationwide, we draw a random, measured sample of polling stations. A field officer at each selected station photographs or records the result announced there (the official count form) and sends it to a data center. From that sample we estimate the result for the whole population of polling stations, together with a margin of error.

Notice the difference from an ordinary opinion survey. An opinion survey asks people about their choice or intention. A quick count records the actual result at the polling station, so it does not depend on respondents’ honesty or memory. That is why a well-designed and well-run quick count is usually very accurate, often within one percentage point of the official result.

It also differs from an exit poll, which interviews voters as they leave the polling station. An exit poll still relies on what voters say (and some prefer not to tell the truth), while a quick count reads the official numbers at the station.

Why can a small sample be so accurate?

Because inferential statistics is built for exactly that. If the sample is random and large enough, the sample estimate lands close to the population value, with an error we can calculate. Think of tasting a pot of soup: one spoonful, properly stirred, is enough to know the whole pot. In a quick count, the “stirring” is the random selection of stations, usually with stratification (for example by region) so that every group is represented.

Five reasons a quick count can miss

A quick count can be spot on, but sometimes it is not. These are the most common causes.

1. A sample of stations that is not truly random

If stations are chosen because they are “easy to reach” or because a field officer happens to be available there, the sample is biased toward places that are easy to access (usually urban areas and main roads). The result leans toward the voting pattern of those areas. A documented random selection is the most basic requirement.

2. An outdated sampling frame

The list of stations used for selection must match reality on election day. If stations were merged, split, or moved, or the list is simply old, some areas have no chance of being selected, and the sample no longer represents the real population.

3. Incoming data that is incomplete and unbalanced

Data from stations does not arrive all at once. Urban stations with good signal usually report first, while remote stations report later. If an organization announces results too early, when only part of the data is in and it is dominated by certain areas, the estimate can lean the wrong way. That is why professional firms wait until the data is large and balanced enough before releasing.

4. A very thin margin

This is not an error but a natural limit of the method. If candidate A gets 50.3% and candidate B 49.7%, the gap is 0.6 points. A quick count’s margin of error is typically around one percentage point or more. In that situation, a quick count cannot declare a winner with high confidence. An honest firm will say the result is “still within the margin of error”. A firm that insists on naming a winner is acting beyond the limits of its statistics.

5. Transmission and data entry errors

A blurry photo of a form, a misread number, or a typo are operational risks. Layered verification, including consistency checks (for example, valid plus invalid votes must match the total number of voters present), helps catch these errors before they damage the estimate.

The margin of error is part of the result

The most common mistake in quick count reporting: the result is written as a single number, and the margin of error is forgotten or buried in a footnote. A properly stated quick count result looks like this:

> Candidate A: 54.2% (±1.0%). Candidate B: 45.8% (±1.0%).

This means that, with 95% confidence, candidate A’s true value lies between roughly 53.2% and 55.2%. It is clear that A leads, because A’s lower bound (53.2%) is far above B’s upper bound (46.8%). But if the result is 50.4% (±1.0%) versus 49.6% (±1.0%), the ranges overlap, and the honest conclusion is “cannot be determined yet”.

What makes a quick count trustworthy?

A few signs you can check yourself as a reader:

• The methodology is open: sample size, sampling method, margin of error, and confidence level are stated.
• The sample is documented and can be inspected, not just “we took 2,000 polling stations”.
• The data source is clear: official results at the polling station, not perceptions.
• The organization is transparent about its funding and affiliations.
• It does not force a conclusion when the gap is within the margin of error.
• It states differences from the official result, including when they exist, and explains why.

How quick count relates to the official count

One point needs clarifying: a quick count is not a replacement for the official count by the election authority. It serves as an independent verification tool and as early information for the public. If the quick count and the official result differ greatly, the question should go in two directions: is there a weakness in the quick count methodology, or a problem in the official counting process? Accusing one side without checking the data is equally unscientific.

Lessons that apply beyond elections

The thinking behind a quick count applies to any survey: random sampling, an up-to-date frame, complete data, and honesty about uncertainty. That is why I always say elections are simply the most visible stage for the same statistical principles you use in your thesis or customer survey.

Why tight precision needs a big sample

In a simple random survey, a margin of error of ±1% at 95% confidence requires about 9,600 respondents (under the most cautious assumption of a 50% proportion). A quick count can work with fewer stations because stations are stratified using data from previous elections, and each station carries a large amount of vote information. But the basic principle stays: the tighter the promised precision, the bigger and neater the sample must be, and the more the quality of data transmission matters. If an organization promises very tight precision with a very small sample, ask politely about its methodology.

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