The date of acquiring data should have not have a negative effect on the data, it should continue to be equally representative of the parent population. Spelling of words will be data sample, mainly due to the problem of recognising uncommon names and terminology specific to a certain situation.
I am conscious that incorrect spelling will have an effect on the data, joining of words appears to be worryingly consistent in some newspapers. Typos (typing errors) are certainly inevitable and are a familiar part of newspapers, therefore I can consider them as a valid part of the sample. However, I wish they checked their spelling before sending to press!
Variables
I have chosen the mean characters per word as the X-axis. I feel the mean word length is easier to calculate than the sentence length, making possible predictions of the sentence length easier to obtain. In addition, from looking at the source data, the range of values for word length appear to be proportionally smaller and values more consistent than sentence length. Therefore the following can be stated:
X = Mean word length throughout text
Y = Mean sentence length throughout text
The sample data is listed in ascending order of X, to highlight any correlations that may lie within X and Y values. Characters per word and words per sentences are used throughout.
Data
Mean Values
The mean of the values taken from the table can be used to draw a line of best fit on a scatter diagram. The mean values of X and Y are as follows:
Scatter Diagram
From the spread of the data in the scatter diagram, there appears to be a positive correlation with variables X and Y increasing in concurrence. The distribution of the data is elliptical, shown by the ellipse superimposed on the spread of the greater part of the data. Many of the points lie close to the line of best fit and within the narrow ellipse, indicating a positive correlation between X and Y. In positive sectors (top right; bottom left of mean centre) there is a total of 54 data points, while there are only 4 data points in the negative sectors (top left, bottom right of mean centre), this indicates a good positive correlation.
Normal Distribution
To show data values follow a distribution, the data for the variables X and Y will be independently plotted against a normal distribution curve.
The variables of X and Y both can be approximated to a normal distribution curve. The mean word length distribution is slighted negatively skewed and the mean sentence length appears to be bimodal, with no values equalling 19. In the context of mean sentence length this can be regarded at as quirk in the results, despite the large sample set (unless, for some reason, reporters purposely avoid this number!). Mean word and sentence length are independent random variables – simply replacing short words with long words in a sentence doesn’t make a sentence longer. Both data sets follow a normal distribution curve and the scatter diagram shows the data points can be represented using a line of best fit, therefore the Pearson’s Product Moment correlation coefficient method can be used to show any correlations.
Correlation Coefficient
The Pearson’s Product Moment correlation coefficient can be calculated using the covariance of X and Y.
Covariance = SXY = = 7289.3 – 5.084 × 24.42 = 1.5234
Correlation Coefficient = r = = = 0.9362
As the correlation coefficient is close the 1, the correlation is strong between X and Y independent random variables.
Hypothesis Testing
To test if there is a significant correlation in the representative sample data, I will perform a hypothesis test on the Product Moment Coefficient, r. The hypothesis are as follows:
H0 Null Hypothesis: p = 0 If significant, there no positive correlation present between variables X and Y.
H1 Alternative Hypothesis: p > 0 If significant, there is a positive correlation between random variables X and Y.
On consultation of the tables, the critical value a single tailed test at 2.5% significance, where n = 58, the critical value p = 0.2589. As r: 0.9362 > 0.2589, H0 is rejected. The difference between the critical value and the correlation coefficient is great, this indicates a very strong correlation between the two data samples.
As the alternative hypothesis (H1) is accepted, my premise is correct. As strong correlation can be found between word and sentence length for news articles in the Independent newspaper. I am confident that these correlations lie with other newspapers and in other media such as in magazines, novels, websites and students’ coursework. However, data analysis taken from a larger population indicates that varying styles of writing have an effect on the correlation correlation.
Line of Regression
The line of regression will show differences between the two linear functions used to conjecture data values from another. The sum of squares of X = S X2 = 1509.5 and as the covariance was found as: 1.5234, the following equation can be used to calculate the line of regress in terms of X.
The histogram shows the line of best fit and the line of regression is shown in blue. The equation of the line of regress is: Y = 4.829X – 0.1266. It cuts through the mean at (5.084, 24.46), visible from (19.189, 4) to (28.847, 6). The line of regression crosses the X-axis at -0.1266 and the . The line of regression crosses the X-axis at and the Y-axis at -0.1266.
The line of regression can now be used to estimate a value for X or Y given that the other value is already known. As there is a good correlation it can should give an easier alternative if words or characters were being counted manually.
In my keenness for retrieving and processing articles from the Independent website, I over estimated the amount of data that I needed. A recent article omitted from the same was titled: "French Small Farmers Head For Extinction", the mean characters per word is: 5.12. Using the line of regression: 4.829(5.12) – 0.1266 = 24.6. Percentage error = (100 × 0.1) ÷ 24.7 = 0.4% error. This error is surprisingly small showing a good representation of the scatter diagram data points.
Conclusions
In terms of headline news from the online Independent newspaper, there are several connections between the manner in which prose are written. The implementations of correlations with in the sentence and word structure could indicate on the training reporters need in order to report for the newspaper, or a consistent editorial team. This investigation has shown a clear pattern in an organisations writing style, possibly caused by the influences people have on each other within the newspaper offices. It might even be possible to calculate lines of regressions for word and sentence length of other newspapers. It would be a useful guide to identifying torn-off articles. Similar processes have been used in the past, notably on the works of Shakespeare to find out if he actually wrote everything under his name. Even if statistics are drawn up on sentence structure, there is not enough evidence here to show that correlations equally apply to other organisations or individual writers.
A possible reason for why both variables increase together could be confidence of a writer or the audience to which the article is aimed at.
The relationship between the length of words and sentences has been established within a certain writing style. If a reporter wishes to produce an article which is quick and easy to read information, the sentence and word length will be shorter. Placing full stops at regular intervals quickens the pace of the article, making it appear as if the reader was seeing events happen through their own eyes. A accurate example of this would be "Paddington Train Disaster: ‘I Can See Her Trapped, Her Face Behind Glass, Crying For Help. I Don't Think She Made It’". This article was dialogue heavy account of the disaster, in which a survivor of the crash was speaking of her experiences. The reporter took the opportunity to produce a fast paced article, hopefully making it possible for readers to understand the situation better.
An example of very much slower paced article, where long words and sentences were used, would be: "Workaholic Boss Forced To Take A Holiday Died Saving His Wife In Tour Death Crash". This is another sad story looking at a bitter-ironic tale of an over worked man going on holiday; dying whilst attempting to save his wife. At a glance, the article contains more long adjectives and pronouns than previously.
Evaluation
As with all sampling methods, inconsistencies and errors are almost guaranteed to give inaccuracies within the data. I feel the choice of newspaper was sufficient in comparison with inconsistence newspaper websites like the Guardian and The Times, both which are related to other newspapers (Observer and Sun respectively). In addition the Independent is one of the few newspapers that is not politically biased. The online version of the newspaper is quite likely to be edited earlier than articles for the paper version. This is due to the nature of the web – post the news quickly, otherwise people will read elsewhere. Its could almost be compared to a text version of the news on television. In most instances, editors have time to check over articles before they are sent to press. The paper version is also more final than the web, if a mistake is made once the printing has started there is no way to change it. Some people also rate the web less highly than the traditional paper version, this could all mean a misrepresentation of the newspapers style.
Copying of data from the browser to Word might have resulted in errors. It is possible that some articles were not completely copied. The grammar check process might have miscalculated words if they were incorrectly spelt. This would also lead to inaccuracies.
In order to further improve relevance with the parent population I could investigate all articles written by the Independent newspaper, within a set time scale [of a week]. To expand this investigation and include comparisons between other newspapers it would be sensible to increase the parent population to incorporate other newspapers, with different writing styles. Perhaps it would also be interesting to study articles, organised by subject matter rather than source. This would encompass more texts and could help further improve whether there is a certain style specific to a type of profession.