But there is another kind of distribu on — be er known as dispersion — which shows how a variable is dispersed/spread with respect to its central tendency. DATA VISUALIZATION 2) Rela onship Trees grow taller as they get older in the early years. That's a rela onship between two variables — height and age. height = f(age) In another example, the price of a house depends on the number of beds, number of bathrooms, loca on, square footage etc. This is a rela onship between one dependent and many explanatory variables. price = f(beds, baths, loca on, area) DATA VISUALIZATION: distribu on, rela onship, composi on, comparison The principles of data visualiza on Centuries ago scien sts didn't have cameras to take pictures of distant galaxies or ny bacteria under the microscope. Drawings were the primary medium to communicate observa ons, ideas and even theories. In fact, the ability to draw abstract ideas and objects was an essen al skill for scien sts. We cannot take pictures of the distribu on variables or their correla on. Instead, we communicate them through drawings and illustra ons — aka data visualiza on — through modern tools and technologies. A dataset contains one or more variables, and we can visualize each and their interac on with others in mul ple ways. Which visualiza on to choose depends on data and the kind of informa on we want to communicate. However, fundamentally, they are of four different kinds: Distribu on of a single variable Rela onship between two variables Composi on of a single or mul ple variables Comparison between different categories/individuals 1 Distribu on An important concept in sta s cs and data science is distribu on. Distribu on generally refers to the probability of occurrence of an outcome. In a distribu on of 100 coin flips how many will get heads and how many tails? Frequency distribu ons like this are presented in histograms or curves. If you look at a dataset just as numbers, there is no way to iden fy these rela onships. But in fact, you can, without going into complex sta s cal analysis, with the help of a good visualiza on. Rela onship between age and height of a class of students 3) Comparison The third cornerstone of data visualiza on is Comparison. This kind of visual material compares mul ple variables in datasets or mul ple categories within a single variable. Let's check out the following two visuals: Visuals to show comparisons The one on le compares a variable (salary) between two groups of observa ons (scien sts vs lawyers) on a bar chart. The right panel is also a comparison chart — in this case, comparing a variable (GDP)between two groups (UK and Canada) but along a me dimension. 4) Composi on Have you heard about stacked bar charts? But I'm sure you know what a pie chart is. The purpose of these charts is to show the composi on of one or more variables in absolute numbers and in normalized forms (e.g. percentage). Composi on charts are some of the old school visualiza on techniques that nowadays have limited use cases (do you really need a pie chart to show a composi on of yellow 10% and red 15%?). Nevertheless, some mes they can present informa on in a visually aesthe c and familiar, vintage fashion. Composi on plots: Stacked bar chart (le ) and pie chart (right) Priori ze Your Data Analysis to Achieve Long Term Success Real meaning of an “Insight” As an analyst it is important to priori ze your analysis to have long term success. We see majority of our analysts and data scien sts busy with repor ng and data prepara on. And they get minimal me to provide solid fresh insight which can show a business impact. Proper insight should have followings Analysis and insight Recommended ac on Predicted impact What can I do with an insight? Each analysis should start with a business problem otherwise you will end-up nowhere with your effort. In parallel to that you should know exact areas where you can add value to the business. Below are the 3 ares you can provide recommenda on and make an impact to the business with marke ng analy cs Campaigns recommenda on UI/UX recommenda on Product recommenda ons Filter Your Analysis Requirements to explain a framework to manage your me or mul ply it. ( How To Mul ply Your Time). This framework can be adopted to priori ze your analy cs work to become analy cs ninja not a person who do data puking. Below are the 3 key filters you can used to pick an analysis. Urgency : How soon does it ma er — If the required analysis me sensi ve then you have to priori ze it. E.g. Campaign is running for two weeks and campaign team about to spend $20,000 within 14 days. Proper site search / landing page analysis could save them $5000 within the same week. Jump on it an do it because that opportunity is not going to come again. Importance : How much does it ma er — Campaign team wanted to do a ribu on analysis to improve budget alloca on on mul channel campaigns. In this case, do analysis and provide recommenda on to op mize exis ng campaigns based on a ribu on models. Which can mul ply the return on ads. Significance : How long does it ma er — If there is an analysis which can help you to achieve sustainable growth and save/make money every month then do it. E.g. GA analysis done to iden fy trending product to market could be a good analysis where you can use every week/month/year. Read the complete guide to do that analysis “Google Analy cs Guide to Finding E commerce Products with Higher Demand”. These are analysis, you don't see huge impact immediately but it can provide cumula ve impact in the long run. Way to scale your analysis Eliminate ad-hoc and repe ve repor ng work which can't add value or unable to take ac on to make an impact to the business. Some me you can't avoid the report because your HIPPO wanted that report weekly/daily basis. If you really can't avoid a report then move in to second step and automate the report. Ini ally you might have to spend me on automa ng it but it will save me for you in the long run. Apart from that, you can add a recommenda on along with the automated report to provide few ac on items. Assume that you are sending a daily revenue report but that report shows last week revenue has a drop of 20%, this is a good opportunity for you to show your analy cs skills and find the problem with the revenue report. Once you have the analysis which can provide the recommenda ons to an eCommerce customer, simply write a R code where you can plug to any customer and derive the insight within few mins. Even though you automate it, you require human eye to go through and verify that the report is relevant to the business also it would be handy if you can add comments to increase the value of your report. Comment should come with a person who understand dynamics in the business strategy, products and service and bit of psychology. Finally train people to read a report or do an analysis considering business strategy, products and service, and user psychology. What make you a successful analyst Below process help analyst to climb the ladder quickly. You can do 100 analysis and come up with recommenda on but all these recommenda ons are hypothesis unless you experiment and quan fy the business impact. Start with the problem > Analysis > Ac ons Items > Experiment > Learning You should analyse all the experiment you have done regardless of the results. There are opportuni es even with a loosing experiment. Once you have significant no of experiments under your belt, you will get the ripple effect and the results are extraordinary. Search Data, Trends & Analy cs: Catching the Pulse of a Market and its Consumers. The benefits of search data go beyond SEO-focused keyword research. It is a valuable resource for both market and product research, too. In fact, it can even be integrated into all research types and intelligence analyses where listening to and decoding the consumer's voice plays an important role. WHY SEARCH DATA CAN BE USED AS A KEY SOURCE OF INSIGHT AT THE STRATEGY LEVEL. Why Search Data It would have been prac cally impossible for me to reach the number of analy cs professionals Krista had in her study, so I decided instead to use publicly available search data and see if any noteworthy trends would emerge. At that point in me I was actually already thinking about the poten al benefit of this type of analysis — but for a different reason. Covid-19 was at its peak, and it was becoming obvious that the pandemic was strongly impac ng consumer behaviour. Finding ways to look closer at how consumers search for products online and which new search pa erns are emerging could offer useful insights. On the other hand, Fishtown Analy cs (another term with close-to-meteoric rise) released a popular open-source tool that contributed to its rapid growth. 2. Rising stars in analy cs It's easy to keep the focus only on the terms and topics that have the largest volumes of traffic. However, valuable insights can be found anywhere within the keyword universe. A number of analy cs so ware vendors — in some cases, we could think of them as niche players — have grown very consistently, even though some literally started from zero. An interes ng insight is that three out of the six vendors in the graph belong to the category of 'privacy-first' analy cs, which seems to be a reac on to all the privacy legisla on and discussions that followed over the course of the last few years. It is a strong signal that the new privacy landscape can create significant opportuni es for new players in the analy cs space. 3. Fastest-growing terms in Analy cs jobs and training Moreover with increasing barriers in the tracking of user behaviour, any addi onal data sources that echo the consumer's voice, become highly useful. I hypothesised that search data could also provide another complementary angle to the original analy cs trends research. The Analy cs Trends Research Method I created and data-mined a keyword universe for Analy cs, looking for the fastest-growing trends. The method deployed was as follows. · Create a keyword universe with 30,000 keywords containing the term 'analy cs' using mul ple data sources for the keyword sugges ons. · Retrieve the search volumes from Google for a period of four years. · Apply sta s cal analysis to spot the key trends. 3 Trends in the Analy cs Industry Based on Search Data A number of noteworthy findings naturally emerged from this analysis, which gives a good indica on of the insights that search analysis can produce. These are the fastest-growing analy cs tools, technology and vendors in mainstream analy cs tech. Some of the tools, like Einstein and Spo fy analy cs, have grown in very consistent ways over the course of the last four years. Others, like Google Analy cs 4, follow a hockey s ck curve. It's o en worth looking at the causes of the specific search behaviour for the various topics — in Google Analy cs 4, for example, the surge in search interest was the result of a new product announcement. It's easy to keep the focus only on the terms and topics that have the largest volumes of traffic. However, valuable insights can be found anywhere within the keyword universe. A number of analy cs so ware vendors — in some cases, we could think of them as niche players — have grown very consistently, even though some literally started from zero. This last trend illustrates the increasing demand for analy cs-related roles, such as analy cs engineers and analy cs directors. Demand for specialised roles is followed by strong demand for training and courses that qualify professionals to work in these roles. The high demand for analy cs jobs and educa on is not a surprise in itself, but the pace of growth is quite impressive — look at the rise of search interest for analy cs engineers and analy cs bootcamps. Search Data: Going from Tac cal to Strategic I've used search and keyword data for a long me, mainly working with it from a tac cal standpoint for keyword bidding, ROAS calcula ons, keyword rank tracking and keyword research for SEO purposes. While working on this analysis, it became obvious to me that search data can play a strategic role when making business decisions involving products, markets and consumers. The analysis above only scratches the surface, there is so much more to be explored when the terms within a 'keyword universe' start to be grouped or clustered based on their seman c, lexicographical or other case-specific a ributes. I have priori sed the use and analysis of search data in my work since the analy cs trends study. Below, you'll find several ideas on how to use search data strategically in your organisa on. Ways to Use Search Data Strategically in Your Organisa on 1. Use it to observe the bigger picture of what's changing in an industry. The example in this ar cle comes from the analy cs industry, but it could be any other industry or market. Search data analy cs can be useful in iden fying rising or declining search interest in products, markets or topics, while also helping businesses understand the specific pa erns in which the change unrolls. Some terms, for example, can have a meteoric rise and fall compared to others that grow slowly but surely over me. Understanding these pa erns is essen al when deciding whether to invest in a new product or market for the long term or take advantage of short-term opportuni es, i.e. marke ng campaigns that focus temporarily on topics that spike in search interest for just a short me. DATA VISUALIZATION: distribu on, rela onship, composi on, comparison The principles of data visualiza on Centuries ago scien sts didn't have cameras to take pictures of distant galaxies or ny bacteria under the microscope. Drawings were the primary medium to communicate observa ons, ideas and even theories. In fact, the ability to draw abstract ideas and objects was an essen al skill for scien sts. We cannot take pictures of the distribu on variables or their correla on. Instead, we communicate them through drawings and illustra ons — aka data visualiza on — through modern tools and technologies. A dataset contains one or more variables, and we can visualize each and their interac on with others in mul ple ways. Which visualiza on to choose depends on data and the kind of informa on we want to communicate. However, fundamentally, they are of four different kinds: Distribu on of a single variable Rela onship between two variables Composi on of a single or mul ple variables Comparison between different categories/individuals 1 Distribu on An important concept in sta s cs and data science is distribu on. Distribu on generally refers to the probability of occurrence of an outcome. In a distribu on of 100 coin flips how many will get heads and how many tails? Frequency distribu ons like this are presented in histograms or curves. DATA QUALITY WORKS DATAQWORKS DATAQWORKS@GMAIL.COM