Last updated on: Suppose a bank wants to know whether customers in Mumbai save more on average than customers in Pune. Or a college wants to check if students taught with a new method score differently from those taught the traditional way. In both cases, you are not comparing one group against a fixed number; you are comparing two groups against each other. The challenge is that any difference you observe in your samples might be real, or it might just be the result of random chance. Statistical inference gives us a structured way to settle this question, and the most common tools for the job are the z-test and the t-test . This post walks through when and how to use each one. Table of Contents Why we test for differences between two samples Setting up the hypotheses When the population variance is known: the z-test When is the z-test actually used? When the population variance is unknown: the t-test The pooled variance approach When variances are not equal: Welch’s t-test Checking the assumptions Choosing between the z-test and the t-test A quick worked example Common mistakes to avoid Why we test for differences between two samples 🔗 When researchers compare two groups, they usually focus on the mean , or average, of each group. This could be the average savings of customers in two cities, the average yield of a crop under two fertilizers, or the average recovery time of patients on two treatments. The numbers from your samples will almost always differ to some degree. The real question is whether that difference is meaningful. The central problem is this: are the differences observed between the two sample means genuinely reflecting a difference in the wider populations, or could they simply have arisen from random variation? A sample is only a slice of the population, and two random slices will rarely produce identical averages even when the populations are the same. Hypothesis testing is the method that helps us decide which explanation is more believable. Setting up the hypotheses 🔗 Every test of this kind begins with two competing statements. The null hypothesis (H₀) assumes there is no difference between the two population means, written as μ₁ = μ₂. The null hypothesis is usually a statement of no difference or no relationship . The alternative hypothesis (H₁) is the opposite claim, that the two means are not equal (μ₁ ≠ μ₂). The alternative is really your research suspicion, and the null is its logical opposite. We then assume the null hypothesis is true and ask how surprising our data would be under that assumption. To make a decision, we fix a significance level , denoted α, usually set at 0.05 or 5%. This is the probability we are willing to accept of wrongly rejecting a true null hypothesis, which is called a Type I error . Setting α before collecting data keeps us honest, because we are not tempted to pick a cut-off based on the result we hope to see. After computing a test statistic, we calculate a p-value , which is the probability of observing a difference as large as ours (or larger) if the null hypothesis were actually true. The decision rule is simple: if the p-value is less than or equal to α, we reject the null hypothesis ; if it is larger, we do not. A small p-value tells us the observed difference is unlikely to be the product of chance alone. When the population variance is known: the z-test 🔗 The first scenario is when you already know the variances (or standard deviations) of the two populations. In this case, the appropriate tool is the two-sample z-test . The z-test is used to determine whether there is a significant difference between two means when the population standard deviation is known . It is built on the standard normal distribution , the familiar bell curve. The logic is straightforward. We take the difference between the two sample means, then divide it by the standard error of that difference. The standard error captures how much the difference between sample means is expected to bounce around due to sampling. The resulting z-statistic tells us how many standard errors apart the two means are. A larger z-value points to a bigger, more convincing difference. The formula for the test statistic looks like this: z = (x̄₁ − x̄₂) / √(σ₁²/n₁ + σ₂²/n₂) Here x̄₁ and x̄₂ are the two sample means, σ₁² and σ₂² are the known population variances, and n₁ and n₂ are the two sample sizes. Once we compute z, we compare it against a critical value from the normal distribution. For a two-tailed test at α = 0.05, that critical value is roughly ±1.96. If our z falls beyond this boundary, the difference is statistically significant. When is the z-test actually used? 🔗 The honest answer is: not very often in practice. Knowing the true population variance is rare, because if we cannot even pin down the population mean, we are unlikely to know its variance. In real research, the z-test is generally suitable when the population standard deviation is genuinely known or when the sample sizes are large. The z-test is appropriate when the sample size is large (n ≥ 30) and the population variance is known , and in both situations the data should be approximately normally distributed. Despite its limited real-world use, the z-test is usually the first test students learn. Its results are easy to calculate by hand, it relies only on the standard normal distribution, and it does not require the concept of degrees of freedom. This makes it an excellent stepping stone before moving on to the more practical t-test. When the population variance is unknown: the t-test 🔗 In the vast majority of real situations, we do not know the population variances. We only have the variances calculated from our samples. This is where the two-sample t-test takes over. The two-sample t-test is a method used to test whether the unknown population means of two groups are equal . It uses sample-based estimates of variability instead of known population values. The t-test is built on the t-distribution , which looks similar to the normal curve but has heavier tails. These fatter tails account for the extra uncertainty that comes from estimating variance from a sample rather than knowing it outright. This adjustment matters most when sample sizes are small, which is exactly when the t-test earns its keep. As sample sizes grow, z and t test results converge, and for sample sizes greater than 30 the differences between them become small . The pooled variance approach 🔗 When we can reasonably assume the two populations have equal variances, we combine the two sample variances into a single estimate called the pooled variance . This weights each sample variance by its degrees of freedom, producing an unbiased estimate of the common population variance. The pooled standard deviation is calculated from both sample variances using the formula: S²ₚ = [(n₁ − 1)s₁² + (n₂ − 1)s₂²] / (n₁ + n₂ − 2) The t-statistic then becomes: t = (x̄₁ − x̄₂) / [Sₚ √(1/n₁ + 1/n₂)] This test statistic follows a t-distribution with n₁ + n₂ − 2 degrees of freedom under the null hypothesis . Degrees of freedom essentially count the number of values that are free to vary in the calculation. We use this number to look up the critical t-value from a t-table, against which our computed t is compared. When variances are not equal: Welch’s t-test 🔗 The pooled approach assumes the two populations share the same variance. When that assumption is doubtful, the pooled method is no longer appropriate. In such cases researchers turn to Welch’s t-test , which does not assume equal variances and instead adjusts the degrees of freedom to compensate. Its test statistic divides the difference in means by the square root of the sum of each sample variance divided by its own sample size. Because deciding whether variances are truly equal can be tricky, many researchers rely on Welch’s t when comparing two means . Checking the assumptions 🔗 A t-test is only valid when certain conditions hold. The samples should be independent, the populations should be approximately normally distributed, and, for the pooled version, the population variances should be equal . Independence means the two groups are not linked. If your two samples are actually paired, such as the same students tested before and after a training programme, you would need a paired t-test instead of the independent two-sample version. Choosing between the z-test and the t-test 🔗 The decision really comes down to two questions. First, do you know the population variance? If yes, the z-test is theoretically correct. If no, the t-test is your tool. Second, how big is your sample? With large samples the two tests give nearly identical answers, but with small samples the t-test’s heavier tails give a more honest, more conservative result. In short, z-tests require you to know the population standard deviation, while t-tests use a sample estimate of it . Since the population standard deviation is almost never known in genuine research, the t-test is the workhorse you will reach for most often, whether you are comparing customer savings across two cities, exam scores across two classrooms, or yields across two farming methods. A quick worked example 🔗 Imagine a researcher comparing the average monthly savings of bank customers in two cities. A sample of 40 customers from City A shows a mean saving of ₹5,200, while a sample of 35 customers from City B shows a mean of ₹4,800. The population variances are unknown, so the t-test applies. The researcher computes the pooled variance, calculates the t-statistic using the difference of ₹400, and compares it to the critical t-value at α = 0.05 with n₁ + n₂ − 2 = 73 degrees of freedom. If the calculated t exceeds the critical value (and the p-value falls below 0.05), the researcher concludes that the savings difference between the two cities is statistically significant and unlikely to be due to chance alone. Common mistakes to avoid 🔗 A few errors trip up students regularly. Using a z-test when the population variance is unknown and the sample is small overstates your confidence. Forgetting to check whether the samples are independent or paired leads to the wrong test entirely. And misreading the p-value is widespread: a p-value does not tell you the probability that the null hypothesis is true. It only tells you how likely your data would be if the null were true. Keeping this distinction clear will keep your conclusions sound. Choosing the right test, stating your hypotheses clearly, and interpreting the p-value correctly are the three habits that separate a careful analysis from a misleading one. Once these become second nature, comparing two groups becomes a reliable and powerful part of your research toolkit. What do you think? If you were studying whether a new teaching method genuinely improves student performance, how would you decide between a z-test and a t-test for your data? And how comfortable are you that a “statistically significant” result always reflects a difference that matters in the real world? How useful was this post? Click on a star to rate it! Average rating 0 / 5. Vote count: 0 No votes so far! Be the first to rate this post. We are sorry that this post was not useful for you! Let us improve this post! Tell us how we can improve this post? https://www.jmp.com/en/statistics-knowledge-portal/inferential-statistics/hypothesis-testing https://www.scribbr.com/statistics/type-i-and-type-ii-errors/ https://online.stat.psu.edu/statprogram/reviews/statistical-concepts/hypothesis-testing/p-value-approach https://statisticsbyjim.com/hypothesis-testing/z-test/ https://www.datacamp.com/tutorial/t-test-vs-z-test https://www.jmp.com/en/statistics-knowledge-portal/t-test/two-sample-t-test https://stats.libretexts.org/Bookshelves/Applied_Statistics/Natural_Resources_Biometrics_(Kiernan)/04:_Inferences_about_the_Differences_of_Two_Populations/4.02:_Pooled_Two-sampled_t-test_(Assuming_Equal_Variances) https://www.sparkl.me/learn/as-a-level/mathematics-further-9231/pooled-variance-and-two-sample-comparisons/revision-notes/5155 https://www.almabetter.com/bytes/tutorials/applied-statistics/two-sample-t-and-z-test Leave a Reply Cancel reply Research Methodology 1 Research- Meaning, Concept, Need Definition of Research Need for and Purpose of Research Conceptual Framework of Research and Terminology Basic and Applied Research Scientific Method Research Design Value of Studying Research Methodology for Library and Information Professionals 2 Historical Research Historical Research Definitions What is Not Historical Research? What Constitutes Historical Research? Advantages Limitations Purposes Scope of Application Types Importance in LIS Process of Conducting Research Internet Scientific Research Problems 3 Survey Research Data Collection Sampling How to Conduct a Survey Problems Instruments of Survey Research 4 Experimental Research Experimentation Hypothesis Research Procedure Validity Design of the Experiment Limitations of Experimental Method 5 Fundamental, Applied and Action Research Scientific Method S. R. Ranganathan’s Spiral of Scientific Method Basic Research Applied Research Action Research Descriptive Research Comparative Research Exploratory Research Diagnostic Research Social Research 6 Measurement of Variables Types of Variables Measurement of Qualitative Data Census versus Sample Survey Sampling Procedure Types of Sampling 7 Data Presentation Preparation of a Table Tabular Presentation Graphical Presentation Bar Diagrams Pie Chart 8 Statistical Techniques Measures of Central Tendency Arithmetic Mean Median Mode Measures of Dispersion Variance and Standard Deviation Coefficient of Variation Correlation Pearson’s Product Moment Correlation Regression Analysis Linear Regression Non-linear Regression Time Series Analysis 9 Statistical Packages Statistical Packages Microsoft Excel SPSS Other Software for Statistical Analysis 10 Observation Method Meaning and Definition Purpose Characteristics Planning and Process of Observation Recording of Data Types Advantages and Disadvantages Application in Libraries and Information Centers 11 Questionnaire Method Questionnaire Method: Definition Questionnaire Construction Types of Questionnaires Types of Questions Use of Scales Precautions in Questionnaire Construction Pretesting of Questionnaire Distribution of Questionnaire Response Rate Advantages and Limitations of Questionnaire Method 12 Interview Method Introduction Interviewing Types of Interviews Structured Interview Unstructured Interview Focussed Interview Non-directive Interview Clinical Interview Telephonic Interview Computer Assisted Telephone Interviewing (CATI) System Interview Process Advantages and Limitations of Interview Method 13 Experimental Method Research Problem Appropriate for an Experiment Parts of an Experiment Steps in Planning an Experimental Research Laboratory Experiment and Field Experiment Experimental Research Design Advantages, Disadvantages, and Limitations of Experimental Method 14 Case Study Case Study: Definition, Characteristics, and Importance Uses, Advantages, Disadvantages, and Limitations of Case Study Research Problem Appropriate for a Case Study Research Design in Case Study Steps in Case Study Method Case Study vs Case Work and Other Methods 15 Research Design What is Research Design? Need and Purpose Functions of Research Design Types of Research Design Based on Nature of Investigation Based on Data Collection Methods Based on Number of Contacts Made with the Subjects Based on Reference Period Summary 16 Research Plan Definition Need and Purpose Functions Types Structure Funding Monitoring Ethics 17 Statistical Inference Concept of Statistical Inference Statistical Estimation Concept of Hypothesis Testing Critical Regions and Types of Errors Testing of Hypothesis for a Single Sample Test for Difference between Two Samples Contingency Table 18 Presentation of Results Research Reports and their Types Importance and Significance of Research Reports Preparation of a Research Proposal Research Reports: Plan Outline, Format and Contents Preparation and Organisation of Research Notes Drafting of Research Reports Language and Grammar Physical Production 1 Research- Meaning, Concept, Need Definition of Research Need for and Purpose of Research Conceptual Framework of Research and Terminology Basic and Applied Research Scientific Method Research Design Value of Studying Research Methodology for Library and Information Professionals 2 Historical Research Historical Research What is Not Historical Research? What Constitutes Historical Research? Scope of Application Importance in LIS Process of Conducting Research Scientific Research 3 Survey Research Data Collection How to Conduct a Survey Instruments of Survey Research 4 Experimental Research Experimentation Research Procedure Design of the Experiment Limitations of Experimental Method 5 Fundamental, Applied and Action Research Scientific Method S. R. Ranganathan’s Spiral of Scientific Method Applied Research Action Research

Descriptive Research Comparative Research Exploratory Research Diagnostic Research Social Research 6 Measurement of Variables Types of Variables Measurement of Qualitative Data Census versus Sample Survey Sampling Procedure Types of Sampling 7 Data Presentation Preparation of a Table Tabular Presentation Graphical Presentation 8 Statistical Techniques Measures of Central Tendency Arithmetic Mean Measures of Dispersion Variance and Standard Deviation Coefficient of Variation Pearson’s Product Moment Correlation Regression Analysis Linear Regression Non-linear Regression Time Series Analysis 9 Statistical Packages Statistical Packages Microsoft Excel Other Software for Statistical Analysis 10 Observation Method Meaning and Definition Characteristics Planning and Process of Observation Recording of Data Advantages and Disadvantages Application in Libraries and Information Centers 11 Questionnaire Method Questionnaire Method: Definition Questionnaire Construction Types of Questionnaires Types of Questions Precautions in Questionnaire Construction Pretesting of Questionnaire Distribution of Questionnaire Advantages and Limitations of Questionnaire Method 12 Interview Method Types of Interviews Structured Interview Unstructured Interview Focussed Interview Non-directive Interview Clinical Interview Telephonic Interview Computer Assisted Telephone Interviewing (CATI) System Interview Process Advantages and Limitations of Interview Method 13 Experimental Method Research Problem Appropriate for an Experiment Parts of an Experiment Steps in Planning an Experimental Research Laboratory Experiment and Field Experiment Experimental Research Design Advantages, Disadvantages, and Limitations of Experimental Method Case Study: Definition, Characteristics, and Importance Uses, Advantages, Disadvantages, and Limitations of Case Study Research Problem Appropriate for a Case Study Research Design in Case Study Steps in Case Study Method Case Study vs Case Work and Other Methods 15 Research Design What is Research Design? Need and Purpose Functions of Research Design Types of Research Design Based on Nature of Investigation Based on Data Collection Methods Based on Number of Contacts Made with the Subjects Based on Reference Period 16 Research Plan Need and Purpose 17 Statistical Inference Concept of Statistical Inference Statistical Estimation Concept of Hypothesis Testing Critical Regions and Types of Errors Testing of Hypothesis for a Single Sample Test for Difference between Two Samples Contingency Table 18 Presentation of Results Research Reports and their Types Importance and Significance of Research Reports Preparation of a Research Proposal Research Reports: Plan Outline, Format and Contents Preparation and Organisation of Research Notes Drafting of Research Reports Language and Grammar Physical Production Share on Mastodon