LSE Data Analysis M6U1 Notes

© 202

5

LSE All Rights Reserved

MODULE 6 UNIT 1 Regression analysis

Download to read ad-free

© 202

5

LSE All Rights Reserved getsmarter.com

|

info@getsmarter.com +44 203 457 5774 (UK)

|

+1 224 249 3522 (US)

|

+27 21 447 7565 (SA) Page 2 of 25

lse.ac.uk

Table of contents

1.Introduction3

2.Correlation3

2.1 Types of correlation5

2.2 Causality7

3.Regression analysis8

3.1 Variables8

3.2 Summary statistics9

3.3 Correlation coefficient11

3.4 Regression line11

3.5 Analysis of variance (ANOVA)14

3.6 Multivariate analysis16

4.The challenges of linearity17

5.Examples17

5.1 Bivariate analysis18

5.2 Multivariate analysis20

6.Conclusion24

7.Bibliography24

Unlock this document
Upload a document to download this document or subscribe to read and download.
or
Unlock the next 2 pages after an ad


© 202

5


|


|

+1 224 249 3522 (US)

|

+27 21 447 7565 (SA) Page 3 of 25

lse.ac.uk

Learning outcomes: LO1:

Recognise the fundamentals of regression analysis.

LO2:

Discuss the complexity of linearity and its suitability to analysing data.

1.Introduction

In February 1922, the LSE adopted its coat of arms along with the Latin motto “

rerum cognoscere causas

”, the English translation of which is “to understand the causes of things” (LSE, 2014). Like the LSE, regression analysis is concerned with

understanding

the reasons behind phenomena. What are some of the "causes of things” a data analyst will encounter and investigate? You could ask the question: what causes an increase in sales? It may be an increase in marketing expenditure, perhaps a change in season, or maybe a public holiday. What drives customer satisfaction? It could be great service, affordable prices, or company reputation. More likely, it is a combination of these factors; but which one has the greatest influence?

Correlation

and

regression

allow you to see the connection between the values of two or more measurable variables. When two variables are analysed, this process is referred to as a

bivariate analysis

. The analysis of more than two variables is called a

multivariate analysis

. Correlation is how you measure the strength of a linear relationship, while regression is how you represent that linear relationship. The potential difficulty of achieving this linearity is briefly discussed in Section 4. In this set of notes, you will explore the fundamental components of regression analysis for continuous variables. In Unit 2, you will be introduced to the added difficulty of performing regression analysis on categorical variables (those with binary outcomes, such as "yes" or "no"). To compare the relationship between two variables, you will require a set of values in paired form.

2.Correlation

Paired data

are any two variables that can be compared with each other or share a common other variable; for example, the number of ice creams sold in a day and the number of sunscreen bottles sold in a day. Both are dependent on seasonality, or the time of year, and not necessarily on one another. However, these two variables are expected to have some form of correlation. Consider the following example of a paired dataset, where the number of people unemployed in certain parts of a city is compared to the number of low-cost housing options in the same parts of that city.

or


© 202

5


|


|

+1 224 249 3522 (US)

|

+27 21 447 7565 (SA) Page 4 of 25

lse.ac.uk

Table 1:

Number of people unemployed and number of low-cost houses in various parts of a city.

Number of unemployed

2,655 1,500 1,489 2,148 3,158 1,780 1,211 1,958 2,478

Number of low-cost houses

920 452 395 756 1,179 524 300 720 898 The first step in determining the relationship between these two variables is to create a

scatterplot

(sometimes called a scatter diagram). This allows you to visually inspect whether the variables are related. If you plot the values provided in Table 1, with the number of people unemployed on the

x

-axis and the number of low-cost houses on the

y

-axis, you get the scatterplot shown in Figure 1.

Figure 1:

A scatterplot of the number of low-cost houses and people unemployed.

Figure 1 gives the impression that there is some sort of linear relationship between the number of people unemployed and the number of low-cost houses in the various parts of the city. The scatterplot shows that, as the number of people unemployed increases, the number of low-cost houses also increases, almost forming a straight line. However, the relationship is not perfectly linear, as the points do not lie on a straight line, and a certain amount of scatter exists.


© 202

5


|


|

+1 224 249 3522 (US)

|

+27 21 447 7565 (SA) Page 5 of 25

lse.ac.uk

2.1 Types of correlation

When one variable is expected to increase as the other increases, as in Figure 1, it is called

positive correlation

. Inversely, if one variable is shown to decrease as the other increases, it is known as

negative correlation

. When two variables show no signs of linear correlation, they are

uncorrelated

. This means that there is no linear relationship between the two variables. However, it does not necessarily mean that there is no relationship; the variables might be exponentially or quadratically related, for example. These types of correlation will not be covered in this course – only linear correlation will be addressed. Figures 2 to 4 presents some examples of correlation. Figure 2 is an example of positively-correlated data, where an increase in marketing expenditure has resulted in an increase in sales (note that while you are only considering correlation at the moment, you could plausibly infer causality, i.e. marketing expenditure affects total sales, rather than vice versa).

Figure 2:

Total sales in relation to marketing expenditure – positive correlation.


© 202

5


|


|

+1 224 249 3522 (US)

|

+27 21 447 7565 (SA) Page 6 of 25

lse.ac.uk

Figure 3 shows the negative correlation between the total sales of a product and its price.

Figure 3:

Total sales in relation to price – negative correlation.

Figure 4 shows two variables, namely customers’ shoe sizes and their monthly income, which have no evident correlation. Thus, you could conclude that they are uncorrelated. However, it is possible that they are correlated in some other way, just not linearly; although, based on Figure 4, there is no obvious pattern in the scatter.


© 202

5


|


|

+1 224 249 3522 (US)

|

+27 21 447 7565 (SA) Page 7 of 25

lse.ac.uk

Figure 4:

Income per month in relation to shoe size – no correlation.

Correlation is measured as a value between -1 and 1. This value is called the

correlation coefficient

. A value of -1 indicates a strong,

negative

, linear relationship between the

x

and

y

variables. Similarly, a value of 1 indicates a strong,

positive

, linear relationship between

x

and

y

. A correlation coefficient of 0 indicates that

x

and

y

are

uncorrelated

, meaning they are not linearly related. The calculation of the correlation coefficient by hand is explored in Section 3 of these notes. The correlation coefficient calculation in Microsoft Excel, and the interpretation thereof, will be discussed in Unit 3 of this module. You will often hear statements made about the correlation between variables, but how can you be certain of such stated relationships? Consider the following variables and the correlation they are expected to have.

Table 2:

Variables and their expected correlation.

Variables Expected correlation

The height and weight of a person Positive correlation Number of hours of sunshine and rainfall Negative correlation Ice cream sales and sunscreen sales Positive correlation Time spent studying and test grades Positive correlation Education and income Positive correlation Absence from work and productivity Negative correlation Petrol consumption and ice cream sales No correlation


© 202

5


|


|

+1 224 249 3522 (US)

|

+27 21 447 7565 (SA) Page 8 of 25

lse.ac.uk

It is important to note that a correlation existing between variables does not necessarily mean that they are dependent on one another, or that one’s values are the cause of the other's.

2.2 Causality

When two variables have a linear relationship, you may wonder if they also have a

causal

relationship. Suppose the following three examples of variables have been shown to have strong correlations: 1.Number of calories eaten and weight gained.2.A country’s GDP and average life expectancy of its citizens.3.Average salary of teachers and alcohol consumption by country.Would it be reasonable to think that there is a

causal

relationship between these variables? 1.It would be perfectly reasonable to conclude that a person’s weight is influencedby their diet and, as a result, that a person is likely to gain more weight on a high-calorie diet. Therefore, you could conclude that a causal correlation exists betweennumber of calories eaten and weight gained.2.It is tricky to establish if GDP is responsible for life expectancy, so it is difficult toestablish a causal relationship. However, it could be argued that richer countrieshave a population that lives longer because their citizens have access to moreadvanced healthcare and other basic services, meaning that there could be acausal relationship between a country’s GDP and the average life expectancy of its citizens.3.It would be frivolous to conclude that increasing the remuneration of teacherswould increase alcohol consumption. It is much more plausible that both variablesare the result of various factors – including the country’s economy and its socialvalues – and probably do not have a causal relationship.It should be clear, at this point, that you should take care when interpreting correlated relationships, as

correlation does not imply causality

.

Explore further:

Tyler Vigen humorously emphasises that correlation does not equal causality on his website Spurious Correlations, which shows correlations between unrelated variables.


© 202

5


|


|

+1 224 249 3522 (US)

|

+27 21 447 7565 (SA) Page 9 of 25

lse.ac.uk

3.Regression analysis

Once a linear relationship has been identified between two variables by assessing the correlation coefficient, a regression analysis can be performed to explain this relationship. There are three main reasons a regression analysis would be performed:

●

To establish and interpret the unknown variables in a known relationship

●

To understand the reasons for such a relationship and determine causality

●

To forecast the response of one variable because of another

3.1 Variables

In a simple linear regression model, two variables are present: the

dependent (or response) variable,

and the

independent (or explanatory) variable

.

●

y

is the dependent (or response) variable, meaning the factor that you are trying toexplain.

●

x

is the independent (or explanatory) variable, meaning the factor that is thoughtto have an influence on

y

.In the earlier example of the number of people unemployed and the corresponding number of low-cost houses, it could be argued that unemployment is what drives the need for low-cost housing. Using this argument, unemployment levels would be the explanatory variable which explains the response variable, low-cost housing.

3.2 Summary statistics

In order to draw inferences from the outcome of regression analyses, some summary statistics need to be defined first. For a paired dataset – where the linear relationship has been determined and the dependent and independent variables have been defined – the corrected sum of squares and products need to be calculated.

Note:

These summary statistics are automatically calculated when performing a regression analysis using built-in functions, such as those found in Microsoft Excel. However, in order to understand how they are calculated, and to interpret them, their equations are discussed in this section. The sum of squares is calculated using the squared distance that each data point is situated away from the mean of that variable. In other words, how far away any

x

value is from

x̄

, or how far away any

y

value is from

ȳ

.


© 202

5


|


|

+1 224 249 3522 (US)

|

+27 21 447 7565 (SA) Page 10 of 25

lse.ac.uk

The sum of the squares of these values is known as the

corrected sum of squares for

x

-data (

S

xx

)

and the

corrected sum of squares for

y

-data (

S

yy

)

. Another summary statistic is the

corrected sum of cross-products of

x

-data and

y

-data (

S

xy

).

This is calculated by summing the products of the distance the

x

-value is from its mean with the distance the

y

-value is from its mean.

Example 1:

Consider again the following paired dataset, where the number of people unemployed was positively correlated to the number of low-cost houses in various parts of a city. The number of low-cost houses has been defined as the response variable, as it could be argued that it is dependent on unemployment.

Table 3:

Unemployment and reported low-cost housing data.

Number of unemployed (

x

)

2,655 1,500 1,489 2,148 3,158 1,780 1,211 1,958 2,478

Number of low-cost houses (

y

)

920 452 395 756 1,179 524 300 720 898 The means of the

x-

data and

y

-data can be calculated as

x̄

= 2,041.89 and

ȳ

= 682.67, respectively. The corrected sums of squares and corrected sum of cross-products can then be calculated.

Table 4:

Corrected sum of squares and cross-products.

x y

(

x

-

x̄

)

2

(

y

-

ȳ

)

2

(

x

-

x̄

)(

y

-

ȳ

)

2655 920 375,905.23 56,327.11 145,511.70 1,500 452 293,643.57 53,207.11 124,995.70 1,489 395 305,686.12 82,752.11 159,047.70


© 202

5


|


|

+1 224 249 3522 (US)

|

+27 21 447 7565 (SA) Page 11 of 25

lse.ac.uk

2,148 756 11,259.57 5,377.78 7,781.48 3,158 1,179 1,245,704.01 246,346.78 553,963.15 1,780 524 68,585.79 25,175.11 41,553.04 1,211 300 690,376.35 146,433.78 317,953.48 1,958 720 7,037.35 1,393.78 -3,131.852,478 898 190,192.90 46,368.44 93,909.26

Σ

3,188,390.89 663,382.00 1,441,583.67

These calculated values will now be used in further calculations to determine the correlation and regression components.

3.3 Correlation coefficient

Having calculated the summary statistics of

S

xx

,

S

yy

, and

S

xy,

these can be used to determine the correlation coefficient,

r

. Recall from Section 2.1 that the correlation coefficient is a value between -1 and 1, indicating the strength of the linear relationship between two measurable variables. The equation for the correlation coefficient is:

Example 2:

Looking once more at the unemployment and low-cost housing data in the previous example, where

S

xx

= 3,188,390.89,

S

yy

= 663,382.00, and

S

xy

= 1,441,583.67, the correlation coefficient can be calculated: With

r

being very close to positive 1 (at 0.991), it can be said that there is a strong, positive, linear relationship between the number of people unemployed and the number of low-cost houses in various parts of the city.

3.4 Regression line

Let’s summarise the different variables that have been defined up to this point.

●

x

= the independent (or explanatory) variable. This variable is not dependent onanother and can be changed or chosen as needed. The mean of

x

is

x̄

.

●

y

= the dependent (or response) variable. This variable is expected to change asthe explanatory variable changes. The mean of

y

is

ȳ

.


© 202

5


|


|

+1 224 249 3522 (US)

|

+27 21 447 7565 (SA) Page 12 of 25

lse.ac.uk

●

S

xx

= corrected sum of squares for

x

-data. This value is calculated by summing thesquares of the distances of all the various values of

x

in relation to

x̄

.

●

S

yy

= corrected sum of squares for

y

-data. This value is calculated by summing thesquares of the distances of all the various values of

y

in relation to

ȳ

.

●

S

xy

= corrected sum of cross-products for

x

-data and

y

-data.

●

r

= the correlation coefficient. A value between -1 and 1, which indicates whether there is a positive or negative linear relationship between the

x

and

y

variables, aswell as the strength of the relationship.For two variables, when a correlation has been identified, the nature of their relationship can be represented using an estimated regression line. As with any straight line, the equation for the estimated regression line uses the standard linear equation of a

y

-intercept summed with the product of the gradient and the

x

-value. In this case, the

y

-intercept is defined as

β

0

and the gradient as

β

1

. This produces the equation: Where

β

0

and

β

1

are unknown parameters that are estimated using the following equations: As this is an approximation of the true relationship between the explanatory and response variables, each

y

observation requires some random digression from the initial approximated line. Each observation of

y

almost lies on the approximated line, with a certain offset distance. This perturbation is called the

error term

and is denoted with the Greek letter epsilon,

ε

. Therefore, the basic model is modified to include the error term, and becomes: The error terms corresponding to the data points are assumed to be independent and identically normally distributed, with a mean of 0 and a constant, unknown variance. This means that any of the individual error terms has a high probability of being small and relatively close to the regression line.


© 202

5


|


|

+1 224 249 3522 (US)

|

+27 21 447 7565 (SA) Page 13 of 25

lse.ac.uk

Figure 5:

Normally distributed error terms.

Example 3:

Consider once more the data regarding the number of people unemployed and the related number of low-cost houses, and recall the calculated summary statistics of

S

xx

= 3,188,390.89,

S

yy

= 663,382.00, and

S

xy

= 1,441,583.67. Earlier on, the means were calculated as

x̄

= 2,041.89 and

ȳ

= 682.67. Substituting these values into the beta equations, the equation for the estimated regression line can be calculated: Substituting the beta-hat values, the estimated regression line equation becomes: When the

x

and

y

variables are plotted (view Figure 1 for a reminder) and a trendline is included in Excel, the graph shown in Figure 6 is generated.


© 202

5


|


|

+1 224 249 3522 (US)

|

+27 21 447 7565 (SA) Page 14 of 25

lse.ac.uk

Figure 6:

Regression line created in Excel for low-cost houses and unemployment.

The regression line is useful in that it contains the

beta coefficient

for a paired dataset, also known as the gradient of the line. It only indicates the best-fitting straight line, and the variance in the dataset should be determined and analysed next. Note that the calculated regression line paves the way for predictions made from regression models. These methods of prediction are discussed in Module 7.

3.5 Analysis of variance (ANOVA)

The overall objective of using regression analysis is to explain the response variable as a function of the explanatory variable. For example, for every unit increase spent on marketing, what will the increase in sales figures be? More specifically, the goal is to explain the

variation

in the response variable. When performing a simple linear regression, this is done by using a single explanatory variable. The total variation in the response variable for a paired dataset is the corrected sum of squares of the

y

-data: This can now be called the

total sum of squares

(

TSS

). The total sum of squares can be split into two parts: the first is the part of the

TSS

that can be explained using the model,


© 202

5


|


|

+1 224 249 3522 (US)

|

+27 21 447 7565 (SA) Page 15 of 25

lse.ac.uk

called the

explained sum of squares

(

ESS

); and the second, which cannot be explained, called the

residual sum of squares

(

RSS

).

TSS = ESS + RSS

Figure 7 shows how the total variation consists of the explained variation and the residual variation.

Figure 7:

Explained and residual variations as parts of total variation.

The

TSS

will thus be the explained sum of squares combined with the residual sum of squares to form the total sum of squares. The equation for calculating the

ESS

is: The overall fit of the regression model can be expressed as the proportion of the total variability in the response variable that is explained by the model. In other words, how much of the total sum of squares is contained in the explained sum of squares? This is known as the

coefficient of determination,

R

2

.


© 202

5


|


|

+1 224 249 3522 (US)

|

+27 21 447 7565 (SA) Page 16 of 25

lse.ac.uk

As

R

2

is a proportion, it is a value between 0 and 1, and the closer

R

2

is to 1 the better the

explanatory power of the model

.

Note:

Although residuals and error terms are closely related and easily confused, take note of the following fundamental difference between them.

●

Error term: A "correction" term to reflect the fact that the true relationship between

x

and

y

is not exactly linear. It measures the difference between the observed valueand the true value.

●

Residual: The difference between an observed value of

y

and its fitted value fromthe estimated regression line.

Example 4:

Continuing once more with the example of the number of people unemployed and the related number of low-cost houses, recall the calculated summary statistics of

S

xx

= 3,188,390.89,

S

yy

= 663,382.00, and

S

xy

= 1,441,583.67. Calculating

R

2

: This means that 98.25% of the total variation in the number of low-cost houses can be explained by unemployment. Note that this is extremely high, indicating that the model is a great fit.

3.6 Multivariate analysis

Multivariate analysis is used when a response variable is expected to have a linear relationship with multiple explanatory variables. These relationships are not usually equal, as one explanatory variable can have a greater influence on the response variable than another. For two explanatory variables, the resultant linear regression line equation obtains another factor, becoming: As more explanatory variables are included, more factors are added to the equation.


© 202

5


|


|

+1 224 249 3522 (US)

|

+27 21 447 7565 (SA) Page 17 of 25

lse.ac.uk

4.The challenges of linearity

Not all paired datasets are created equal. Remember that for regression to achieve valid and reliable results, the relationship between the explanatory and response variables needs to be at least approximately linear. This is not always the case for relationships between variables. Take, for example, the two variables plotted on a scatter diagram in Figure 8. In this case, the relationship between them is represented with the equation

y = x(1−x)

for values of

x

between 0 and 1.

Figure 8:

Non-linear variables.

Clearly this relationship is not linear. In fact, it follows a parabolic curve. Although not linear, a well-defined, non-linear relationship still clearly exists between the two variables, so they are certainly not independent. For interest's sake, the correlation coefficient for this pair is basically 0, at 5.37 × 10

-17

, resulting in an uncorrelated outcome. It is often useful to explore other ways of representing the same data, only in a more linear form. By transforming an exponential dataset into its logarithmic equivalent, it is transformed into a more linear dataset that does fit the assumptions needed for regression analysis. Regression analysis relies on assumptions made about the linearity of a paired dataset. In Unit 2, you will delve deeper into what these assumptions are and how you can ensure that your regression analyses present valid and reliable results by investigating this linearity.


© 202

5


|


|

+1 224 249 3522 (US)

|

+27 21 447 7565 (SA) Page 18 of 25

lse.ac.uk

5.Examples

Consider two examples of regression analysis, the first with two variables (bivariate analysis), and the second with multiple variables (multivariate analysis.)

5.1 Bivariate analysis

Working as a marketing manager, Simone is asked to determine whether the amount the company spent on marketing in the last year was related to the total number of sales generated by the company. She is also asked to investigate the nature of the relationship, should one exist. Simone obtains the marketing expenditure and total sales figures for the last year, as shown in Table 5.

Table 5:

Marketing expenditure and total sales for 12 months.

Month Marketing expenditure Total sales

January £1,550 £11,874 February £980 £8,123 March £1,225 £8,578 April £845 £7,015 May £1,125 £8,746 June £1,024 £7,210 July £784 £6,758 August £1,358 £10,045 September £1,204 £10,455 October £1,050 £8,987 November £1,402 £11,874 December £905 £6,251 Simone determines from this paired dataset that total sales could be dependent on the marketing expenditure. Therefore, she defines the marketing expenditure as the independent variable,

x

, and total sales as the dependent variable,

y

. A scatterplot of this data indicates that there seems to be a positive linear relationship between these variables, as shown in Figure 9.


© 202

5


|


|

+1 224 249 3522 (US)

|

+27 21 447 7565 (SA) Page 19 of 25

lse.ac.uk

Figure 9:

Total sales as dependent variables of marketing expenditure.

Simone calculates the summary statistics as:

x̄

= £1,121

ȳ

= £8,826

S

xx

= 607,624.00

S

yy

= 40,106,668.67

S

xy

= 4,559,474.00 Using these values, she calculates the correlation coefficient as 0.9236, indicating that there is indeed a strong, positive, linear relationship between marketing expenditure and total sales. The regression line components are calculated as: Substituting the beta coefficient estimates, the estimated regression line equation becomes:


© 202

5


|


|

+1 224 249 3522 (US)

|

+27 21 447 7565 (SA) Page 20 of 25

lse.ac.uk

Plotting the regression line on the scatterplot, Simone generates the graph seen in Figure 10.

Figure 10:

Regression line of total sales in relation to marketing expenditure.

Lastly, Simone calculates the coefficient of determination,

R

2

: Simone reports back to her boss, indicating that the total sales generated by the company can be attributed, in some way, to marketing expenditure. She indicates that 85.31% of the variation in total sales can be explained by marketing expenditure. The high value indicates that the model is a very good fit. She also notes that not all the sales can be explained by marketing expenditure, as is expected, as other factors will also influence how well the company performs.

5.2 Multivariate analysis

Harriet works for a fast-food company and is asked to illustrate the effects of marketing instruments on the weekly sales volumes of a certain product over a three-year period. The marketing instruments used by Harriet’s company were:

●

Changing the promotion price

●

Varying the amount spent on feature advertising


© 202

5


|


|

+1 224 249 3522 (US)

|

+27 21 447 7565 (SA) Page 21 of 25

lse.ac.uk

●

Varying the amount spent on display measuresHarriet decides to plot the sales volumes over time, and generates the chart shown in Figure 11.

Figure 11:

Total sales volumes per week.

The time series plot shows that there is momentum in the data. It also shows that there are major fluctuations in total sales volumes, which might be due to any of the three marketing instruments. Because of the large fluctuation in sales volumes, Harriet decides to plot the logarithmic total sales volumes against the same time series, which generates the chart provided in Figure 12.


© 202

5


|


|

+1 224 249 3522 (US)

|

+27 21 447 7565 (SA) Page 22 of 25

lse.ac.uk

Figure 12:

Logarithmic total sales volumes per week.

Immediately, she notices smaller fluctuations in sales volumes because of a more linear time series. Next, Harriet plots the logarithmic total sales volumes against each of the three marketing instruments. From the chart in Figure 13, Harriet notices that there is a negative correlation between promotion price and logarithmic total sales volumes. This makes logical sense, as an increase in price would normally result in decreased sales.


© 202

5


|


|

+1 224 249 3522 (US)

|

+27 21 447 7565 (SA) Page 23 of 25

lse.ac.uk

Figure 13:

Logarithmic total sales volumes in relation to promotion price.

From the chart in Figure 14, Harriet notices that there is a positive correlation between the amount spent on feature advertising and the logarithmic total sales figures. This also makes logical sense, as an increase in advertising would normally result in increased sales.

Figure 14:

Logarithmic total sales volumes in relation to amount spent on feature advertising.

Lastly, Harriet plots the amount spent on display measures against logarithmic total sales volumes. From the chart in Figure 15, Harriet notices that there is no clear correlation between the logarithmic total sales volumes and the amount spent on display measures.

or


© 202

5


|


|

+1 224 249 3522 (US)

|

+27 21 447 7565 (SA) Page 24 of 25

lse.ac.uk

Figure 15:

Logarithmic total sales volumes in relation to amount spent on display measures.

Explore further:

You can download the data used in this example as an Excel file in the Module 6 downloads folder, should you wish to plot these graphs yourself, or gain further insight from the data.

6.Conclusion

Regression analysis and correlation allow you to investigate the relationship between measurable variables. However, regression analysis presents many challenges, including the appropriateness of linear regression and the assumptions that are made with linearity. Many variables are not linearly correlated and can therefore present inaccurate results when used in regression analysis, unless steps are taken to transform the variables into variables with linear correlations. Understanding the underlying theory in calculating the various components of regression enables you to understand what they represent and how to interpret them. Unit 3 explores the steps involved in performing regression analyses in Microsoft Excel.

or


© 202

5


|


|

+1 224 249 3522 (US)

|

+27 21 447 7565 (SA) Page 25 of 25

lse.ac.uk

7.

Bibliography

Abdey, J. 20

24

. Elementary Statistical Theory [ST102 Lecture material]. Department of Statistics, The London School of Economics and Political Science. London School of Economics and Political Science. 2014.

A history of LSE

. Available: http://www.lse.ac.uk/study/meetLSE/WhyLSE/aHistoryOfLSE.aspx [2018, September 17].

or

LSE Data Analysis M6U1 Notes

Uploaded by

LSE Data Analysis M6U1 Notes

Uploaded by

Share this document

You might also like